Rasa Pro Change Log

All notable changes to Rasa Pro will be documented in this page. This product adheres to Semantic Versioning starting with version 3.3 (initial version).

Rasa Pro consists of two deployable artifacts: Rasa Pro and Rasa Pro Services. You can read the change log for both artifacts below.

[3.18.0] - 2026-07-20

Rasa Pro 3.18.0 (2026-07-20)

Deprecations and Removals

• Breaking change: Removed the deprecated LanguageModelFeaturizer NLU component and its internal HuggingFace integration (rasa.nlu.utils.hugging_face). This removes Rasa's direct dependency on the transformers library, which was affected by CVE-2026-1839 (arbitrary code execution via a malicious checkpoint file).

  Assistants whose NLU pipeline references LanguageModelFeaturizer must migrate to a supported featurizer (for example ConveRTFeaturizer, or a CALM pipeline using LLMCommandGenerator) before upgrading. transformers is no longer a first-party component dependency. It remains a transitive dependency of gliner and is now constrained to >=5.0.0 (which contains the fix) wherever it is installed — i.e. on the gliner-based extras (full, pii). The standalone transformers extra and the sentencepiece dependency have also been removed.

Features

• Added a local (in-process) document retrieval backend for the Rasa copilot. The backend chunks documentation pages by heading, fixed-token sliding window, or whole-page strategy; embeds chunks using fastembed (ONNX); and stores vectors in a parquet artifact. At serving time the copilot ingests a prebuilt parquet artifact so no embedding runs on the query path. Vector and hybrid (vector + BM25) search modes are supported.

Improvements

• Removed copy.deepcopy() calls from four processor.py debug-log events (extract.slots, message.parse, actions.log, actions.policy_prediction), cutting unnecessary per-turn CPU overhead on the message-processing hot path.

• Breaking change — model retrain required: Vector stores now use native faiss-cpu, qdrant-client, and pymilvus SDKs instead of LangChain wrappers. At release, MINIMUM_COMPATIBLE_VERSION will be bumped to 3.18.0, so model tarballs trained on earlier Rasa versions are rejected at load time with UnsupportedModelVersionError. This is intentional: retrain is mandatory, not optional.

  Who is affected by the retrain requirement?

  • In-model FAISS used by Enterprise Search, CALM flow retrieval (FlowRetrieval), and IntentlessPolicy. Legacy LangChain-pickled index.pkl artifacts are detected and rejected (the pickle is never opened). New trains write store.json alongside index.faiss.
  • Enterprise Search with FAISS: retrain the assistant and re-ingest documents after upgrade.
  • CALM flow retrieval / IntentlessPolicy with FAISS: retrain the assistant after upgrade.

  What is not affected?

  • Remote Qdrant and Milvus collections — existing ingested vectors and endpoints configuration are unchanged; no re-ingestion required.
  • YAML / endpoints configuration for built-in vector stores — no config migration.
  • Custom InformationRetrieval subclasses that do not depend on LangChain types — continue to work; update constructors to accept a Rasa EmbeddingClient (embed / aembed) instead of LangChain-style embed_query / embed_documents method names.

  Migration steps (summary):

  1. Upgrade Rasa Pro to 3.18.0.
  2. Re-ingest knowledge-base documents if you use FAISS Enterprise Search.
  3. Run rasa train to produce a new model tarball (native store.json format).
  4. Deploy the retrained model.

Bugfixes

• Fixed an issue where custom actions executed directly by a long-running Rasa server retained stale state after action files were modified or deleted. Previously, changes to custom action code were not picked up until the server was restarted. The direct custom action executor now properly clears cached modules, re-imports updated code from disk, and removes orphaned actions that no longer exist.

• rasa test e2e --coverage-report and rasa llm finetune prepare-data now write one YAML per original source file (mirroring the input folder structure) instead of a single flat YAML. This fixes silent fixture corruption — and, since PR #5461, a hard DuplicateFixtureException on reload — when the same fixture name is defined differently across source files (ENG-2647).

  Breaking Change: The --e2e-failed-tests flag now takes a directory instead of a file path; failed tests are written into a timestamped subdirectory under it.

• Upgrade langsmith dependency to 0.8.5 and starlette to 1.0.1 to address CVE-2026-45134 and CVE-2026-48710.

• Fixed a leaked privacy-manager background scheduler thread that kept rasa test e2e (and the finetuning and dialogue-understanding test commands) from exiting cleanly.

  The privacy manager's background thread is now stopped on every path that retires an agent — the test CLI commands, server model reload/unload, and builder agent swaps — instead of only on server shutdown.

• Bumped langchain-classic (~1.0.7, fixes CVE-2026-45134) and starlette (~1.3.1, fixes CVE-2026-54283) to resolve high-severity dependency security vulnerabilities. Also raised the langchain-core floor to ~1.3.3, which is required by langchain-classic 1.0.7.

• Hardened deserialization and webhook parsing across the agent runtime so missing or malformed fields no longer crash production with KeyError or similar failures.

  Tracker and dialogue stack: from_dict methods for pattern stack frames, flow stack frames, chit-chat/search frames, and slots now use .get() with safe defaults for fields added after initial release, so older serialized tracker data loads cleanly after upgrades. Event deserialization skips individual malformed events with a warning instead of failing the entire tracker load.

  LLM and command processing: LLM tool-call parsing, command JSON deserialization, and the command processor tolerate missing optional fields and skip invalid entries rather than raising. Flow retrieval no longer assumes document metadata is always present.

  Flows and model loading: Flow step and link from_json deserializers validate required keys up front. Model metadata, graph schema, and local model fingerprint loading handle missing fields with clear errors or safe defaults instead of bare dict access.

  Channels and voice: Slack, Facebook Messenger, VIER CVG, Deepgram (V1), and AudioCodes webhook handlers validate incoming payloads defensively—returning HTTP 400 for invalid interactive Slack messages (including missing user.id), logging and skipping malformed Facebook/Audiocodes events, and raising clear errors for malformed VIER CVG recipient IDs without leaking credential blobs in logs.

  Flow execution: MCP tool step input/output mapping uses safe dict access when mapping fields are absent.

• Removed the unsupported eager_eot_threshold option from the Deepgram Flux (flux-*) ASR config.

• Upgrade:

  • ujson to version 5.13.0 to address CVE-2026-54911
  • cryptography to version 48.0.1 to address GHSA-537c-gmf6-5ccf
  • msgpack to version 1.2.1 to address GHSA-6v7p-g79w-8964
  • pydantic-settings to version 2.14.2 to address GHSA-4xgf-cpjx-pc3j
  • aiohttp to version 3.14.1 to address CVE-2026-34993, CVE-2026-47265, CVE-2026-54273,CVE-2026-54274, CVE-2026-54276, CVE-2026-54277 & CVE-2026-54278

• Fixed rasa test e2e failing on user steps with metadata when tracker events included extra runtime-injected keys (e.g. model_name, default_language).

• Fixed end-to-end testing incorrectly loading eval/conftest.yml (simulation/evaluation config) when resolving E2E conftest.yml files, which could cause e2e test CLI to fail schema validation in projects that also define simulation / eval configuration. E2E conftest discovery now only checks each ancestor directory directly (not nested paths) and ignore <project>/eval/conftest.yml file path at the assistant project root (detected via config.yml, domain.yml, domain/ or config/), so nested eval/conftest.yml files remain valid. Also fixed conftest path lookup hanging indefinitely when no assistant project root was found, which caused rasa test e2e CLI tests to time out.

• Fixed the --credentials argument to default to credentials.yml instead of None, ensuring consistent behavior with other default configuration paths like endpoints.yml and config.yml. Updated the --domain help message to clarify that it defaults to domain.yml or domain/ directory when not specified (this was always the runtime behavior due to validation fallback logic, but the help text was misleading).

• Fixed rasa run actions crashing at Sanic startup with ConstructorError: could not determine a constructor for the tag '!env_var' when endpoints.yml contained an unquoted ${ENV_VAR} reference.

• Fixed empty Input/Output on command-generator LLM spans in Langfuse tracing: Rasa now forces LiteLLM to emit its own request span when Langfuse is configured (LiteLLM 1.87.0 otherwise skips it under a parent span), and exports the prompt and completion text for self-hosted text-completions calls.

• Security patch — upgrade strongly recommended: Rasa Pro 3.18.0 removes the LangChain dependency cluster, which pulled in langsmith with a known security vulnerability (arbitrary local file read via TracingMiddleware)

Miscellaneous internal changes

Miscellaneous internal changes.

[3.17.2] - 2026-07-13

Rasa Pro 3.17.2 (2026-07-13)

Bugfixes

• Fixed end-to-end testing incorrectly loading eval/conftest.yml (simulation/evaluation config) when resolving E2E conftest.yml files, which could cause e2e test CLI to fail schema validation in projects that also define simulation / eval configuration. E2E conftest discovery now only checks each ancestor directory directly (not nested paths) and ignore <project>/eval/conftest.yml file path at the assistant project root (detected via config.yml, domain.yml, domain/ or config/), so nested eval/conftest.yml files remain valid. Also fixed conftest path lookup hanging indefinitely when no assistant project root was found, which caused rasa test e2e CLI tests to time out.
• Restored the boto3 dependency range to >=1.40.60,<1.43 on the 3.17.x line, matching 3.16.15. This range had been widened on main and 3.16.x but was never backported to 3.17.x, which blocked upgrades for users needing newer boto3 versions (e.g. for AWS Bedrock).

Miscellaneous internal changes

Miscellaneous internal changes.

[3.15.31] - 2026-07-15

Rasa Pro 3.15.31 (2026-07-15)

Bugfixes

• Fixed the --credentials argument to default to credentials.yml instead of None, ensuring consistent behavior with other default configuration paths like endpoints.yml and config.yml. Updated the --domain help message to clarify that it defaults to domain.yml or domain/ directory when not specified (this was always the runtime behavior due to validation fallback logic, but the help text was misleading).

Miscellaneous internal changes

Miscellaneous internal changes.

[3.15.30] - 2026-07-07

Rasa Pro 3.15.30 (2026-07-07)

Deprecations and Removals

• Breaking change: Removed the deprecated LanguageModelFeaturizer NLU component and its internal HuggingFace integration (rasa.nlu.utils.hugging_face). This removes Rasa's direct dependency on the transformers library, which was affected by CVE-2026-1839 (arbitrary code execution via a malicious checkpoint file).

  Assistants whose NLU pipeline references LanguageModelFeaturizer must migrate to a supported featurizer (for example ConveRTFeaturizer, or a CALM pipeline using LLMCommandGenerator) before upgrading. transformers is no longer a first-party component dependency. It remains a transitive dependency of gliner and is now constrained to >=5.0.0 (which contains the fix) wherever it is installed — i.e. on the gliner-based extras (full, pii). The standalone transformers extra and the sentencepiece dependency have also been removed.

[3.17.1] - 2026-07-07

Rasa Pro 3.17.1 (2026-07-07)

Deprecations and Removals

• Breaking change: Removed the deprecated LanguageModelFeaturizer NLU component and its internal HuggingFace integration (rasa.nlu.utils.hugging_face). This removes Rasa's direct dependency on the transformers library, which was affected by CVE-2026-1839 (arbitrary code execution via a malicious checkpoint file).

  Assistants whose NLU pipeline references LanguageModelFeaturizer must migrate to a supported featurizer (for example ConveRTFeaturizer, or a CALM pipeline using LLMCommandGenerator) before upgrading. transformers is no longer a first-party component dependency. It remains a transitive dependency of gliner and is now constrained to >=5.0.0 (which contains the fix) wherever it is installed — i.e. on the gliner-based extras (full, pii). The standalone transformers extra and the sentencepiece dependency have also been removed.

Bugfixes

• Hardened deserialization and webhook parsing across the agent runtime so missing or malformed fields no longer crash production with KeyError or similar failures.

  Tracker and dialogue stack: from_dict methods for pattern stack frames, flow stack frames, chit-chat/search frames, and slots now use .get() with safe defaults for fields added after initial release, so older serialized tracker data loads cleanly after upgrades. Event deserialization skips individual malformed events with a warning instead of failing the entire tracker load.

  LLM and command processing: LLM tool-call parsing, command JSON deserialization, and the command processor tolerate missing optional fields and skip invalid entries rather than raising. Flow retrieval no longer assumes document metadata is always present.

  Flows and model loading: Flow step and link from_json deserializers validate required keys up front. Model metadata, graph schema, and local model fingerprint loading handle missing fields with clear errors or safe defaults instead of bare dict access.

  Channels and voice: Slack, Facebook Messenger, VIER CVG, Deepgram (V1), and AudioCodes webhook handlers validate incoming payloads defensively—returning HTTP 400 for invalid interactive Slack messages (including missing user.id), logging and skipping malformed Facebook/Audiocodes events, and raising clear errors for malformed VIER CVG recipient IDs without leaking credential blobs in logs.

  Flow execution: MCP tool step input/output mapping uses safe dict access when mapping fields are absent.

• Fixed action_session_start being incorrectly rejected on legacy trackers that lack a ConversationInactive event. The session-start guard was treating a time-expired session as already active, leaving the first message after a long inactivity period stranded in pattern_code_change instead of starting a fresh session.

• Removed the unsupported eager_eot_threshold option from the Deepgram Flux (flux-*) ASR config.

• Upgrade:

  • ujson to version 5.13.0 to address CVE-2026-54911
  • cryptography to version 48.0.1 to address GHSA-537c-gmf6-5ccf
  • msgpack to version 1.2.1 to address GHSA-6v7p-g79w-8964
  • pydantic-settings to version 2.14.2 to address GHSA-4xgf-cpjx-pc3j
  • aiohttp to version 3.14.1 to address CVE-2026-34993, CVE-2026-47265, CVE-2026-54273,CVE-2026-54274, CVE-2026-54276, CVE-2026-54277 & CVE-2026-54278

• Fixed rasa test e2e failing on user steps with metadata when tracker events included extra runtime-injected keys (e.g. model_name, default_language).

Miscellaneous internal changes

Miscellaneous internal changes.

[3.17.0] - 2026-06-23

Rasa Pro 3.17.0 (2026-06-23)

Features

• Rasa now integrates with Langfuse for LLM observability. Key execution hot-paths — message processing, command generation, agent calls, and MCP tool calls — are instrumented with @observe spans that nest into a single trace per user turn, including bot responses as trace output and per-turn metadata (sender ID, model ID, assistant ID).

• Custom actions can now stream both text responses token-by-token, and rich responses to the end user in real time.

  Action authors initiate the streaming by calling await dispatcher.stream_start(), then call await dispatcher.stream_chunk(text=token) for each partial token and await dispatcher.stream_end() when the response is complete. Rich content (buttons, images, custom JSON) can be streamed mid-response with await dispatcher.stream_chunk(buttons=[...]). On non-streaming transports, stream_end() automatically replays the accumulated chunks as normal utter_message() calls, so actions remain compatible with every channel without code changes.

  Streaming is supported over both the gRPC action server transport and the direct custom action executor (in-process).

• Add a new capabilities API endpoint GET /conversations/<conversation_id>/capabilities and a new built-in capabilities default action action_default_capabilities to enable dynamic "what can you do?" responses without hard-coded flow lists.

  The new endpoint: GET /conversations/<conversation_id>/capabilities returns structured, conversation-aware capabilities metadata for flows (including startability based on current guard evaluation), so custom actions and external integrations can build dynamic "what can you do?" responses without hard-coded flow lists.

  The new default action: action_default_capabilities generates a user-facing capabilities reply directly in Rasa by listing currently startable flows for the active conversation, providing an out-of-the-box implementation for capability discovery. In order to use this action, add it to the user flow handling capabilities user questions, for example:

  flows:
    respond_with_capabilities:
      description: Respond to user questions about what the agent can assist with.
      steps:
        - action: action_default_capabilities

• When an ReAct/A2A agent call or MCP tool call fails, the internal-error pattern frame now receives structured context in its info field instead of only default values with an empty info dict.

  Call sites populate info so you can branch in pattern_internal_error (for example in flow predicates or responses) by failure kind and source. Supported keys include error_source (agent or mcp_tool), agent_name, tool_name, mcp_server, and error_message, plus related agent metadata where applicable (such as flow id from the failing step).

  For example, override the default pattern flow and branch on context.info.error_source:

  pattern_internal_error:
    steps:
      - noop: true
        next:
          - if: context.info.error_source is "mcp_tool"
            then:
              - action: utter_mcp_tool_failed
                next: END
          - if: context.info.error_source is "agent"
            then:
              - action: utter_agent_call_failed
                next: END
          - else:
            - action: utter_internal_error_rasa
              next: END

• When a user interrupts the assistant (voice barge-in) while a custom action is streaming chunks, Rasa Pro now gracefully cancels the stream instead of continuing to deliver audio and accumulate events.

  What happens on interruption:

  • Rasa Pro signals the action server to stop streaming by sending an AckStreamChunks RPC call (gRPC transport) or setting a cancellation flag (direct executor). The action author can react by catching the cancellation inside cancel_stream() on the CollectingDispatcher.
  • All chunks that were already dispatched before the interruption are saved to the tracker store as BotUttered events (tagged with streamed_chunk: true so they are not re-delivered to the output channel).
  • The Deepgram TTS WebSocket session is cleanly closed — including the case where the chunk_end event arrives while already in drain mode — preventing a RuntimeError: Concurrent call to receive() is not allowed error on the next bot utterance.
  • Any events returned by the action in its final_result (e.g. slot sets) are still applied, so the dialogue state remains consistent even when the response was cut short.

  This behaviour applies to both the gRPC action server transport and the direct custom action executor (in-process).

• Added evaluate_agent MCP tool for automated agent evaluation: simulate multi-turn conversations with an LLM-based user, run deterministic assertions (including a new sequencing assertion type), score quality with LLM-as-a-judge using customisable prompt templates, and store per-run results. Evaluation is configured via eval/conftest.yml, scenarios are defined as YAML files in eval/scenarios/ and results are stored in eval/results/<experiment_id>/.

  Added a validate_scenario MCP tool for pre-run YAML validation of scenario structure, assertion schema, slot existence, and slot value type compatibility.

• Added UserBargeIn tracker events for voice conversations, so barge-ins are preserved in the conversation history.

• Added barge-in support for the AudioCodes voice-stream channel by sending AudioCodes speech start and stop events when Rasa detects an interruption.

• Add support for the SignalWire channel for Voice. External URL script can be used for setting up the SignalWire connection. The script can be found on 'host_url/webhooks/signalwire/webhook'.

• Added OutputDeliveryResult for output channels to return delivery signals (events and failure status) from send_response. Voice channels now emit a TTSFinished tracker event containing TTS latency metadata (tts_time_to_first_byte_ms, tts_total_time_ms). A BotTurnEnded tracker event carrying command-processor and prediction-loop execution times is recorded at the end of each bot turn.

• Add support for Vonage channel for voice streaming. When setting up the Vonage application use 'host_url/webhooks/vonage/webhook' as answer url and 'host_url/webhooks/vonage/call_status' as event url.

• Rasa assistants can now act as A2A sub-agents: with an a2a_server block in endpoints.yml, rasa run exposes the Agent2Agent protocol on the same Sanic port as REST and channel webhooks. Orchestrators discover capabilities via AgentCard, send turns over JSON-RPC, and receive A2A task lifecycle updates mapped from Rasa's dialogue state.

  Enable and configure

  Add a2a_server to endpoints.yml. Only description is required; other fields have sensible defaults.

  a2a_server:
    url: "http://localhost:5005"          # optional; inferred from --port / --interface / SSL
    description: "My banking sub-agent"
    include_conversation_repair: true     # false → terminal completed instead of repair input_required
    task_timeout_seconds: 600             # per-task safety net; 0 disables
    a2a_message_cache_ttl_seconds: 600    # messageId replay TTL; defaults to task timeout
    max_contexts: 1000                    # in-memory session cap; 0 disables
    # agent_card_path: ./static-card.json # optional static AgentCard; url still resolved at startup
    # auth: …                             # JWT bearer

  When agent_card_path is omitted, the public AgentCard is auto-generated from user-facing flows at server startup (after the model loads); the advertised URL is resolved from url or from --port / --interface / SSL flags.

  TLS

  A2A uses the same port as the rest of the Rasa server. Configure HTTPS with the usual rasa run flags — not under a2a_server in endpoints.yml.

  Rasa serves HTTPS directly — clients connect to Rasa over TLS using your certificate and key (the same cert covers REST, channels, and A2A):

  rasa run --ssl-certificate /certs/server.pem --ssl-keyfile /certs/server-key.pem

  When url is omitted, the AgentCard URL scheme follows --ssl-certificate. Set an explicit url: "https://..." when the public hostname or port differs from the local bind address.

  HTTPS handled by a reverse proxy or ingress — set url to the public base URL orchestrators use. Rasa can listen on plain HTTP behind the proxy while the AgentCard advertises HTTPS:

  a2a_server:
    description: "My banking sub-agent"
    url: "https://public.example.com"

  Startup guardrails (fail fast with actionable errors):

  • domain.session_config.start_session_after_expiry must be false when A2A is enabled — resumed orchestrator contextId values reuse the same Rasa sender_id; auto session restart would silently reset slots.
  • SANIC_WORKERS=1 is required until Redis-backed A2A task/message stores ship — multiple workers break messageId idempotency, in-flight HTTP 409, orchestrator cancel, and max_contexts. Scale with additional replicas and ingress session affinity instead.

  Session config is validated when a model is loaded with a2a_server in endpoints.yml (e.g. rasa run via load_agent). The worker-count check runs only when starting the Sanic server (rasa run).

  HTTP surface

  Route	Purpose
  POST /	A2A JSON-RPC (message/send, message/stream, tasks/cancel, tasks/get, …)
  GET /.well-known/agent-card.json	Public AgentCard (flow-generated at startup, or static file from agent_card_path)
  GET /.well-known/agent.json	Deprecated alias of the AgentCard

  contextId in A2A requests maps to Rasa sender_id. Each orchestrator message gets a new task_id; input_required keeps the context reserved for follow-up turns on the same contextId.

  Task states and structured status

  Rasa maps dialogue stack and tracker state to A2A task states (working, input_required, completed, failed, canceled, rejected, auth_required). Terminal and interactive-terminal updates include a DataPart with state, active_flow, current slots, and (when applicable) persisted slot values from flows that declare persisted_slots. A TextPart carries the user-visible bot utterance when present so clients reading only status.message still see NLG.

  message/stream emits working status and artifact deltas during token/chunk streaming custom actions; blocking message/send returns the final task.

  Orchestrator operations

  • tasks/cancel — idle cancel on an input_required context, or signal in-flight work; publishes canceled and records FlowCancelled when a user flow was active.
  • messageId deduplication — retries with the same (contextId, messageId) replay the cached terminal task without re-running the turn; a concurrent duplicate while in flight returns HTTP 409 with a JSON-RPC error (retry after the first completes).
  • Session inactivity — ConversationInactive does not release an input_required A2A context; the next message on the same contextId continues the flow when start_session_after_expiry: false (the required A2A setting).

• Added support for Deepgram Flux (flux-*) models in the deepgram ASR engine. Flux uses Deepgram's /v2/listen endpoint with model-integrated end-of-turn detection, replacing silence-timer-based endpointing with lower-latency EOT signals.

  Configure by setting model: flux-general-en (or flux-general-multi) inside language_map. Three optional fields tune EOT behaviour: eot_threshold, eager_eot_threshold, and eot_timeout_ms. Existing Nova (nova-*) configurations are unchanged.

• Added the chirp voice stream channel for Bluejay CHIRP websocket integrations with raw PCM audio frames, optional speech events, Basic authentication, interruption signalling, and L16 24 kHz internal audio support.

• Outbound MCP and A2A calls can now fetch and decrypt credentials only at call time via customer hooks. Hooks return an OutboundCallResult whose metadata Rasa merges into the outbound wire (MCP _meta or A2A Message.metadata). Credentials are never written to slots, the tracker, Kafka, or the inspector — customers store ciphertext in their own secret store keyed by sender_id; Rasa reads via the hook when the call is made.

  Configure hooks

  Flow direct MCP and ReAct external MCP tools — one hook per MCP server in endpoints.yml (flows reference the server by name only; there is no pre_call_hook on flow steps):

  mcp_servers:
    - name: banking_api
      url: http://banking:8000/mcp/
      type: http
      pre_call_hook: my_auth.credentials.resolve_banking_meta
      meta_map:
        static:
          api_version: "v2"

  Static meta_map values are merged first; hook metadata is applied on top (hook wins on key collision).

  ReAct custom tools — override build_custom_tool_call_metadata() on an MCPBaseAgent subclass (default is a no-op).

  A2A outbound — dotted import path in the agent config.yml; stock A2AAgent works without a subclass:

  configuration:
    agent_card: https://payments.example.com/.well-known/agent-card.json
    pre_call_hook: my_auth.credentials.resolve_a2a_meta

  When no hook is configured for a call path, Rasa does not fetch or decrypt credentials and existing behaviour is unchanged.

  Hook contract

  New package: rasa.shared.agents.outbound_call_hook.

  Surface	Context type	Return type
  Flow MCP, ReAct MCP	MCPOutboundCallContext	OutboundCallResult
  ReAct custom tools	MCPOutboundCallContext	OutboundCallResult
  A2A outbound	A2AOutboundCallContext	OutboundCallResult

  Hooks are async def and receive a frozen, read-only context (sender_id is the secret-store lookup key). Rasa also accepts a plain dict return value and normalizes it to OutboundCallResult. Secrets go to metadata only — not tool mapping.input, Message.parts, or top-level A2A context_id / task_id.

  Fail-closed runtime

  Hook exception or invalid return aborts the outbound call: flow MCP routes to pattern_internal_error; ReAct agents return an error result without calling the remote endpoint. Rasa logs metadata key names only, never values.

  Validation

  rasa train / bot validation fails with validation.pre_call_hook.unresolved when a configured pre_call_hook import path cannot be resolved. Agent validation applies the same check for A2A configuration.pre_call_hook.

• Added optional AWS Marketplace license validation at rasa run startup for container deployments sold through AWS Marketplace. When RASA_AWS_MARKETPLACE_ENABLED=true, Rasa performs a PROVISIONAL CheckoutLicense call via AWS License Manager before channels or the Sanic server start; validation failure exits the process so the pod never serves traffic.

  Set RASA_AWS_MARKETPLACE_PRODUCT_ID to the Product ID from the AWS Marketplace Management Portal. The deployment must have IAM credentials that can call License Manager (for example via IRSA on EKS) and an active Marketplace subscription that includes the AWS::Marketplace::Usage entitlement. When the feature is disabled (the default), startup behaviour is unchanged.

  Structured logs record license.marketplace.checkout.verified on success and license.marketplace.checkout.failed on failure, without logging client or consumption tokens.

Improvements

• A2A agent validation now requires TICKET_LOCK_LIFETIME to be at least as long as max_polling_time, preventing the conversation lock from expiring while A2A polling is still in progress.

• Replaced the generic "Start the server from your IDE" message after rasa tools init with IDE-specific instructions for Claude Code, Cursor, VS Code, and JetBrains.

• Added rasa tools status command that displays current project configuration and runtime readiness.

• Propagate the assistant's default language to event metadata under "default_language".

• LLM latency metrics (time-to-first-token and total duration) are now captured and surfaced in BotUttered metadata for the contextual response rephraser and MCP agent.

• Added asynchronous A2A push notification handling for long-running agent tasks. Rasa now persists task status updates on the conversation tracker, can proactively flush pending agent updates through channels that own a live connection, and falls back to delivering the pending update before the next user message when immediate delivery is not available.

• Added ConcurrentKafkaEventBroker, a non-blocking Kafka event broker that offloads publishing to a background thread pool so the event loop is never blocked by Kafka I/O or retry sleeps. Configure it with type: concurrent_kafka in endpoints.yml. It accepts all config options of KafkaEventBroker plus executor_max_workers (default: 1).

• The callback channel supports opt-in streaming responses with stream=true by sending discrete callback events for stream start, text deltas, and stream end.

• Added HTTP push notifications to the A2A server. When an orchestrator supplies a pushNotificationConfig.url on message/send, message/stream, or via tasks/pushNotificationConfig/set, Rasa POSTs task state updates to that callback as the task progresses. message/stream emits intermediate states (submitted, working) as well as the terminal state. The public AgentCard advertises the push_notifications capability when the feature is enabled.

  Push notifications are disabled by default (push_notifications_enabled: false) because callback URLs are client-controlled and can be an SSRF vector when the A2A endpoint is reachable by untrusted callers. When enabled, callback URLs are validated on registration and before each POST: only http/https URLs with a hostname are accepted; loopback, link-local, and private-network targets are rejected (including hostnames that resolve to those addresses via DNS); URLs with embedded credentials are rejected. HTTP redirects are not followed, so a registered public callback cannot redirect into an internal target. Outbound push POSTs use an explicit httpx timeout (5s connect, 30s overall). Callback hostname DNS checks in the URL policy are bounded by a 5s application timeout. Operators may optionally restrict callbacks further with push_notification_allowed_hosts (exact hostname or subdomain match). Rejected URLs surface as InvalidParamsError on the JSON-RPC methods that register them.

  DNS validation at registration and before each POST does not pin the resolved IP for the outbound connection; httpx performs a separate lookup at connect time (DNS rebinding TOCTOU). Recommended deployment controls: keep the feature disabled unless required, use push_notification_allowed_hosts, prefer HTTPS callbacks, and treat callback registration as a trusted orchestrator action.

  Also added an override on RasaA2ARequestHandler to POST the terminal canceled task when a callback URL is registered. The reason for this is that the a2a-sdk supplied on_cancel_task does not send push notifications.

• The A2A sub-agent server now supports optional bearer JWT authentication for orchestrators calling Rasa over the A2A protocol. When a2a_server.auth is set in endpoints.yml, every protected request must include Authorization: Bearer <jwt>; missing or invalid tokens receive HTTP 401 with a WWW-Authenticate: Bearer response header before message processing starts. This applies to the JSON-RPC endpoint (POST /) and AgentCard discovery (GET /.well-known/agent-card.json, including the deprecated alias). If auth is omitted, behaviour is unchanged and no token is required.

  Configure auth under the existing a2a_server block in endpoints.yml. Only type: bearer is supported (API key auth is not available). Use asymmetric verification in production (RS256 + public_key_path to your IdP's PEM public key) or symmetric verification for local/dev setups (HS256/HS512 + secret referenced as an environment variable, e.g. '${JWT_SECRET}' — plaintext secrets are rejected at startup). Optionally set issuer and/or audience to enforce JWT claims.

  Example (production-style):

  a2a_server:
    description: "My domain agent"
    auth:
      type: bearer
      jwt:
        algorithm: RS256
        public_key_path: "/run/secrets/jwt_public_key.pem"
        issuer: "https://auth.example.com"
        audience: "rasa-a2a-agent"

  Orchestrators and local test clients must send the same bearer token when fetching the AgentCard and when calling message/send, message/stream, or other A2A JSON-RPC methods. Auth validates the orchestrator only; end-user identity still belongs in A2A message metadata / slot pre-population, not in this JWT.

• Added configure_litellm_for_rasa() in a lightweight _litellm_config module to centralize LiteLLM setup: it suppresses print-based provider-list debug footers via litellm.suppress_debug_info and sets the LiteLLM logger to WARNING, reducing log noise during LLM, embedding, and router client initialization without pulling in AWS provider dependencies.

• Added orchestrator slot pre-seeding on the inbound A2A server path. On every message/send / message/stream turn, Rasa reads slot context from the incoming A2A Message, applies it as SetSlotCommands before dialogue processing, and strips slot markers from the user text so they are not passed to the LLM.

  Orchestrators can supply slots in any of these shapes (later sources override earlier ones on key conflicts within the payload):

  1. Text transcript — append to the text part:
     • SLOTS: {"slot_name": "value", ...} on its own line, or
     • a trailing fenced block: ```json\n{"slot_name": "value"}\n```
  2. message.metadata — {"slots": {"slot_name": "value", ...}} (merged with JSON-RPC request metadata).
  3. DataPart — {"kind": "data", "data": {"slots": {"slot_name": "value", ...}}} (highest precedence).

  Only slots defined in the assistant domain are written; unknown, builtin, and agent-internal slots are ignored. Values are coerced to the slot type; invalid or out-of-range values (e.g. categorical mismatches) are skipped. String values "none", "null", and "undefined" are stored as null.

  Per-turn precedence: if command generation extracts a SetSlotCommand for a slot from the current user message, that value wins over orchestrator metadata for the same slot on that turn (stale snapshots are ignored). Orchestrator values still pre-fill slots the parser did not set.

  Send an orchestrator slot snapshot on each follow-up message for slots the user did not answer in that turn. Do not rely on user-visible text alone to carry slot values when a collect step is active — prefer metadata.slots or a DataPart, or keep slot JSON out of text the user is answering with.

• Added streaming response support to the SocketIO channel with three new Socket.IO events (stream_start, stream_chunk, stream_end) for real-time token-by-token message delivery from LLM-powered responses. This does not change the behaviour of existing events like bot_uttered, they continue to be sent as before.

Bugfixes

• Fixed MCP task agents exposing set_slot_* tools and listing slots in the task prompt for controlled-only or NLU-only slots in exit_if. Added a slot_mappings_allow_llm_fill helper so tool exposure, prompt rendering, and should_slot_be_set use the same rule for whether the LLM may fill a slot from its mappings.

• Clarified the Copilot system prompt so it explicitly states that users cannot access runtime assistant logs and the Copilot must not ask them to provide those logs.

• Fix training time regression caused by eager component imports.

  Training time increased by ~30-35% (from ~16s to ~22s locally, ~49s to ~68s in CI) after the Beta Agentic PR merge. This was caused by conditional imports of components with external dependencies (TensorFlow, spacy, mitie, etc.) being executed eagerly at module load time during training initialization.

  The fix makes these imports truly lazy by deferring them until the component list is first accessed, eliminating the import overhead and restoring training performance to pre-agentic levels.

• Fix bug when user's message during bot's speech is queued after bot is done speaking. When interruptions are DISABLED, if the bot is speaking and the user talks over it, the bot should ignore that speech entirely. When interruptions are ENABLED, the bot should only stop and listen if the user says enough words to qualify as an interruption (based on the min_words setting).

• Fixed crash when running rasa test on projects containing flows with call: steps that reference agents.

  The rasa test command now correctly initializes the agent registry by calling Configuration.initialise_sub_agents(), aligning its behavior with other CLI commands (rasa train, rasa shell, rasa e2e test, rasa inspect, rasa data).

• Fixed _no_op_observe fallback decorator (used when langfuse is not installed) incorrectly wrapping async functions in a sync wrapper, causing coroutines decorated with @observe to never be awaited.

  Renamed the internal rasa.builder.telemetry.langfuse package to rasa.builder.telemetry.langfuse_integration to avoid shadowing the third-party langfuse package.

• When interruptions are ENABLED, user message should be queued for processing only if it passes interruption criteria.

  If interruptions are disabled, user message should be queued only if user speaks during collect step utterance.

  Updated signature of following methods:

  • send_start_marker: send_start_marker(self, recipient_id: str) => send_start_marker(self, marker_input: MarkerInput)
  • send_intermediate_marker: send_intermediate_marker(self, recipient_id: str) => send_intermediate_marker(self, marker_input: MarkerInput)
  • send_end_marker: send_end_marker(self, recipient_id: str) => send_end_marker(self, marker_input: MarkerInput)
  • send_marker_message: send_marker_message(self, recipient_id: str) => send_marker_message(self, marker_input: MarkerInput)
  • create_marker_message: create_marker_message(self, recipient_id: str) => create_marker_message(self, marker_input: MarkerInput) -> MarkerMessageOutput

  where MarkerInput is a Pydantic-based class:

  class MarkerInput(BaseModel):
      recipient_id: str
      marker_type: MarkerType
      step_type: Optional[StepType] = None

• Remove the "hand over" command from the default GPT-5.1 prompt template. The command had already been removed from all other prompt templates in the 3.16.0 release.

• Fixed a bug where sub-agent flows would end immediately after interruption resumption instead of re-invoking the agent.

  When a flow containing a sub-agent call was interrupted (e.g., by a user digression) and then resumed via pattern_continue_interrupted, the agent frame's state remained INTERRUPTED. This caused the flow executor to skip the agent call and advance to END, triggering pattern_completed prematurely.

  The agent frame state is now correctly reset to WAITING_FOR_INPUT during flow resumption, ensuring the sub-agent is re-invoked with full conversation history as expected.

• Fixed MCP tracing attribute extraction for router-based model groups so tracing no longer crashes with KeyError('provider') when instrumenting ReAct sub-agent send_message calls.

• When interruptions are enabled and bot is not speaking, user utterances should not be checked if they are passing min_words threshold.

• Dialogue Understanding Tests (DUT) now skip rephrase endpoint validation.

  Previously, running DUT would fail with a validation error when the domain contained responses with metadata.rephrase: true as the rephraser is intentionally disabled in DUT.

• Fixed an issue in voice bots where a custom silence_timeout defined on a collect step was reset to the global default after validation failure, causing the re-prompt to use the wrong timeout value.

• Fix an interrupted sub-agent not being properly resumed after the user answers "Yes" to pattern_continue_interrupted. This is fixed by introducing a transient RESUMING agent state that sits between INTERRUPTED and WAITING_FOR_INPUT: resume_flow() now sets the agent frame to RESUMING instead of directly to WAITING_FOR_INPUT; run_agent() detects this and re-invokes the agent with resumed_after_interruption=True.

• Voice stream channels no longer narrate image URLs or attachment strings aloud via TTS. Responses containing an image: or attachment: field are silently dropped on voice channels and a warning is logged instead.

• Fixed utter_corrected_previous_input template to explicitly pair each slot name with its corrected value, preventing LLM rephrasers from swapping the associations when multiple slots are corrected at once.

• Fixed rasa data validate incorrectly rejecting call steps that target subagents or MCP tools when no agent configuration is loaded, instead of emitting a warning.

• Fixed suppress_logs so that concurrent async calls no longer corrupt the process-wide log level, preventing logs from disappearing permanently until pod restart.

• Fixed a spurious "Fatal error on SSL transport" traceback printed to stderr on Windows after rasa test e2e completes by explicitly closing litellm's shared httpx sessions before the event loop shuts down.

• DTMF input collected by the voice channel now bypasses the LLM command generator. The channel injects a SetSlotCommand directly so the target slot can be declared with a controlled mapping instead of from_llm, avoiding an unnecessary LLM round-trip for digit input.

• Fixed rasa data validate defaulting --endpoints to None instead of endpoints.yml, which caused the validator to run with empty endpoints and produce false errors for nlg.type: rephrase setups (ENG-2725).

• Fixed four bugs: fixture dedup in TestSuite.as_writable() now scopes per file (ENG-2647); ask_before_filling slots can be filled when ValidateSlotPatternFlowStackFrame is on the stack (PF-992). Inspector active-flow tile no longer blanks during pattern_user_silence (SWI-1151); Inspector voice latency headline now includes ASR latency and breakdown rows are in chronological order (VO-555).

• Fix the pacing issue in SignalWire channel - when there was a long bot utterance, SignalWire would speed trough generated audio. Fix user barge-in for SignalWire channel.

• Fixed a performance regression in SQLTrackerStore.retrieve() introduced in v3.16, where every retrieve loaded the full conversation history before replay-safe session slicing. Retrieve now fetches a minimum event prefix by trying the most recent session boundaries first (up to three). When reconstruction fails with a dialogue-stack patch error, the store retries from the last stack event in that prefix whose patch applies to an empty stack. If all 3 attempts fail, the store falls back to loading the full conversation history as before. This greatly reduces database I/O and memory use for long-lived conversations while preserving replay-safe behaviour.

• Fixed a crash in rasa-tools where training an assistant with a sub-agent MCP server configured caused a RuntimeError: Attempted to exit cancel scope in a different task than it was entered in, crashing the stdio server and making the next tool call fail with MCP error -32000: Connection closed.

• Removed dead filled_slots_for_active_flow function from rasa/dialogue_understanding/stack/utils.py and its tests; the calling code was replaced in a prior fix but the function was never cleaned up.

• A2A agent run() calls are now properly instrumented with Langfuse, nesting them inside the Rasa message-processing trace instead of generating empty top-level traces.

• Fix two bugs causing random crashes and resource leaks when running Rasa with --workers 2+: a pre-fork event loop left open during JWT setup, and litellm async HTTP clients not being closed on worker shutdown.

  Root cause — Sanic uses fork-based multi-worker mode (legacy=True). create_app() was calling asyncio.new_event_loop() + set_event_loop() in the main process to satisfy sanic-jwt's Initialize(), then leaving the loop open. Forked workers inherited its kqueue/epoll file descriptors, conflicting with each worker's own loop (~50% crash at startup). Separately, close_resources() never called litellm.close_litellm_async_clients(), leaving cached async HTTP clients open when workers exit.

• An explicitly configured api_key in the rasa provider model config now takes precedence over the Rasa license, allowing custom endpoints to authenticate with their own credentials.

• Transient tracker store connection errors (e.g. Postgres restart) no longer flood logs with full stack traces on each retry attempt; the error message is preserved on one line.

• Fix resume_flow() marking the wrong AgentStackFrame as resumed when a flow contains multiple sub-agent frames, which prevented run_agent() from taking its resume branch.

• Resumed sub-agents now send the full conversation history to A2A backends and use stronger directive wording in the resume prompt, preventing the agent from re-asking for information the user already provided before the interruption.

• Fixed multiple Langfuse session IDs being created for a single Dialogue Understanding Test (DUT) run by using the original test-case sender ID as the session ID for all per-step LLM traces.

• Fixed a bug where a mid-conversation language switch (e.g. a custom action setting the language slot followed by utter_language_changed) could silently swallow the confirmation utterance because the call task was torn down by a websockets.exceptions.ConnectionClosedOK ('sent 1000 (OK); then received 1000 (OK)') raised while the ASR engine was being reconnected with the new language.

  The voice ASR and TTS engines now serialize their close+reopen sequence in set_language with an internal asyncio.Lock so concurrent audio producers/consumers can no longer observe a half-closed websocket. The lock also wraps the writer-side I/O methods themselves (send_audio_chunks on ASR; send_text_chunk, signal_text_done, signal_interrupt on the Deepgram/Rime/Azure TTS engines) so they can't race with the reconnect. As a defense in depth:

  • ASREngine.send_audio_chunks now drops a chunk silently if it lands during a reconnect (instead of raising and tearing down the call task).
  • ASREngine.close_connection now nulls out asr_socket so other consumers see "not connected" instead of using a stale, closed reference.
  • The voice channel's _consume_audio_bytes treats ConnectionClosedOK/ConnectionClosedError from the ASR socket as a benign race and continues consuming audio (defense in depth for ASR subclasses that don't swallow ConnectionClosed* internally).

• Refresh the AWS ElastiCache IAM presigned URL before each Redis (re)connect when within 60 seconds of expiry, so lock store and timer store operations no longer fail after >30 minutes of pro-server idleness.

• Upgrading from 3.15.x to 3.16.x no longer causes every subsequent message to fail with InvalidFlowStepIdException when a tracker was mid-flow at upgrade time and a flow step ID was renamed or removed. The flow executor now detects stale step IDs, logs a structured warning, removes the stale frame, and continues processing — rather than propagating the exception as a GraphComponentException.

• Message processing no longer crashes with ValueError: Could not find a user flow frame to cancel when an action fails (for example action_session_start returning HTTP 413) on a session whose residual dialogue stack contains only pattern frames; the internal-error pattern is pushed without attempting to cancel a non-existent user flow.

• Fixed a race condition in KafkaEventBroker where concurrent publish paths (e.g. when using ConcurrentKafkaEventBroker) could create or replace the producer twice. Producer initialisation and reconnect are now protected by a lock.

  Failed connection checks clear the producer so other threads can retry, and reconnect skips producer replacement if another thread already restored the connection.

  _publish now raises when the producer is missing so retry loops cannot exit successfully without delivering the event.

• Updated stale frame detection to use fingerprint comparison so that modified flows are also treated as stale.

• Fixed A2A SSE streaming (message/stream and tasks/resubscribe) so Sanic no longer raises Invalid response type None after response chunks are sent.

• A frame on the dialogue stack that references a step ID no longer present in the model (e.g. after a flow rename across an upgrade) no longer crashes message processing. BaseFlowStackFrame.step() now returns None instead of raising, and _purge_stale_user_flow_frames runs before action_session_start so the action server does not receive stale frames in its tracker payload.

  BREAKING CHANGE: We removed the InvalidFlowStepIdException exception; BaseFlowStackFrame.step() now returns None for stale step IDs instead of raising. If you run into import errors for that exception, you need to adapt your code to handle None instead of catching the exception.

• Fixed voice channels arming the silence timeout before the client finished playing the bot's audio, which could inject a spurious /silence_timeout and advance the flow while the caller was still listening to a long prompt. Silence monitoring now waits until outstanding audio playback markers are acknowledged, barge-ins clear stale markers so monitoring can re-arm, and synchronous channels (e.g. CHIRP) can opt out via tracks_audio_playback.

• Upgrade langsmith dependency to 0.8.5 and starlette to 1.0.1 to address CVE-2026-45134 and CVE-2026-48710.

• Fixed the Inspector's agent name in the top-left corner disappearing on page load or flickering when switching between Preview and Inspect modes.

• Fixed an issue where streaming a response from a custom action recorded each streamed token as a separate BotUttered event, causing the voice inspector transcript to show one bot message per token instead of a single composed message.

  Consecutive streamed text tokens are now aggregated into a single BotUttered tracker event (rich chunks such as buttons or images are still kept separate). On a user barge-in, the partial message streamed before the interruption is preserved as a single event.

• The MCP server now reads flow, domain, and evaluation config files fresh from disk on every tool call instead of returning stale cached content. Previously, edits to these files had no effect until the MCP server was restarted.

Improved Documentation

• When using AWS Bedrock with IAM role-based authentication, the bedrock:InvokeModelWithResponseStream permission is required in addition to bedrock:InvokeModel for any streaming feature, including the Rephraser (nlg.type: rephrase). The Bedrock credential validation error message now explicitly names this permission when IAM auth fails, and the AWS setup documentation has been updated accordingly.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.16.19] - 2026-07-15

Rasa Pro 3.16.19 (2026-07-15)

Bugfixes

• Fixed multiple Langfuse session IDs being created for a single Dialogue Understanding Test (DUT) run by using the original test-case sender ID as the session ID for all per-step LLM traces.
• Fixed rasa test e2e failing on user steps with metadata when tracker events included extra runtime-injected keys (e.g. model_name, default_language).
• Fixed the --credentials argument to default to credentials.yml instead of None, ensuring consistent behavior with other default configuration paths like endpoints.yml and config.yml. Updated the --domain help message to clarify that it defaults to domain.yml or domain/ directory when not specified (this was always the runtime behavior due to validation fallback logic, but the help text was misleading).

Miscellaneous internal changes

Miscellaneous internal changes.

[3.16.18] - 2026-07-07

Rasa Pro 3.16.18 (2026-07-07)

Deprecations and Removals

• Breaking change: Removed the deprecated LanguageModelFeaturizer NLU component and its internal HuggingFace integration (rasa.nlu.utils.hugging_face). This removes Rasa's direct dependency on the transformers library, which was affected by CVE-2026-1839 (arbitrary code execution via a malicious checkpoint file).

  Assistants whose NLU pipeline references LanguageModelFeaturizer must migrate to a supported featurizer (for example ConveRTFeaturizer, or a CALM pipeline using LLMCommandGenerator) before upgrading. transformers is no longer a first-party component dependency. It remains a transitive dependency of gliner and is now constrained to >=5.0.0 (which contains the fix) wherever it is installed — i.e. on the gliner-based extras (full, pii). The standalone transformers extra and the sentencepiece dependency have also been removed.

Bugfixes

• Hardened deserialization and webhook parsing across the agent runtime so missing or malformed fields no longer crash production with KeyError or similar failures.

  Tracker and dialogue stack: from_dict methods for pattern stack frames, flow stack frames, chit-chat/search frames, and slots now use .get() with safe defaults for fields added after initial release, so older serialized tracker data loads cleanly after upgrades. Event deserialization skips individual malformed events with a warning instead of failing the entire tracker load.

  LLM and command processing: LLM tool-call parsing, command JSON deserialization, and the command processor tolerate missing optional fields and skip invalid entries rather than raising. Flow retrieval no longer assumes document metadata is always present.

  Flows and model loading: Flow step and link from_json deserializers validate required keys up front. Model metadata, graph schema, and local model fingerprint loading handle missing fields with clear errors or safe defaults instead of bare dict access.

  Channels and voice: Slack, Facebook Messenger, VIER CVG, Deepgram (V1), and AudioCodes webhook handlers validate incoming payloads defensively—returning HTTP 400 for invalid interactive Slack messages (including missing user.id), logging and skipping malformed Facebook/Audiocodes events, and raising clear errors for malformed VIER CVG recipient IDs without leaking credential blobs in logs.

  Flow execution: MCP tool step input/output mapping uses safe dict access when mapping fields are absent.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.16.17] - 2026-06-24

Rasa Pro 3.16.17 (2026-06-24)

Bugfixes

• Bumped langchain-classic (~1.0.7, fixes CVE-2026-45134) and starlette (~1.3.1, fixes CVE-2026-54283) on 3.16.x to resolve high-severity dependency security vulnerabilities. Also raised the langchain-core floor to ~1.3.3, which is required by langchain-classic 1.0.7.
• Fixed action_session_start being incorrectly rejected on legacy trackers that lack a ConversationInactive event. The session-start guard was treating a time-expired session as already active, leaving the first message after a long inactivity period stranded in pattern_code_change instead of starting a fresh session.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.16.16] - 2026-06-22

Rasa Pro 3.16.16 (2026-06-22)

Bugfixes

• The MCP server now reads flow and domain files fresh from disk on every tool call instead of returning stale cached content. Previously, edits to these files had no effect until the MCP server was restarted.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.16.14] - 2026-06-11

Rasa Pro 3.16.14 (2026-06-11)

Bugfixes

• Fixed MCP task agents exposing set_slot_* tools and listing slots in the task prompt for controlled-only or NLU-only slots in exit_if. Added a slot_mappings_allow_llm_fill helper so tool exposure, prompt rendering, and should_slot_be_set use the same rule for whether the LLM may fill a slot from its mappings.

• Fixed suppress_logs so that concurrent async calls no longer corrupt the process-wide log level, preventing logs from disappearing permanently until pod restart.

• Refresh the AWS ElastiCache IAM presigned URL before each Redis (re)connect when within 60 seconds of expiry, so lock store and timer store operations no longer fail after >30 minutes of pro-server idleness.

• Message processing no longer crashes with ValueError: Could not find a user flow frame to cancel when an action fails (for example action_session_start returning HTTP 413) on a session whose residual dialogue stack contains only pattern frames; the internal-error pattern is pushed without attempting to cancel a non-existent user flow.

• Updated stale frame detection to use fingerprint comparison so that modified flows are also treated as stale.

• A frame on the dialogue stack that references a step ID no longer present in the model (e.g. after a flow rename across an upgrade) no longer crashes message processing. BaseFlowStackFrame.step() now returns None instead of raising, and _purge_stale_user_flow_frames runs before action_session_start so the action server does not receive stale frames in its tracker payload.

  BREAKING CHANGE: We removed the InvalidFlowStepIdException exception; BaseFlowStackFrame.step() now returns None for stale step IDs instead of raising. If you run into import errors for that exception, you need to adapt your code to handle None instead of catching the exception.

[3.16.13] - 2026-06-04

Rasa Pro 3.16.13 (2026-06-04)

Bugfixes

• Fixed a spurious "Fatal error on SSL transport" traceback printed to stderr on Windows after rasa test e2e completes by explicitly closing litellm's shared httpx sessions before the event loop shuts down.

• Fixed a crash in rasa-tools where training an assistant with a sub-agent MCP server configured caused a RuntimeError: Attempted to exit cancel scope in a different task than it was entered in, crashing the stdio server and making the next tool call fail with MCP error -32000: Connection closed.

• Fix two bugs causing random crashes and resource leaks when running Rasa with --workers 2+: a pre-fork event loop left open during JWT setup, and litellm async HTTP clients not being closed on worker shutdown.

  Root cause — Sanic uses fork-based multi-worker mode (legacy=True). create_app() was calling asyncio.new_event_loop() + set_event_loop() in the main process to satisfy sanic-jwt's Initialize(), then leaving the loop open. Forked workers inherited its kqueue/epoll file descriptors, conflicting with each worker's own loop (~50% crash at startup). Separately, close_resources() never called litellm.close_litellm_async_clients(), leaving cached async HTTP clients open when workers exit.

• An explicitly configured api_key in the rasa provider model config now takes precedence over the Rasa license, allowing custom endpoints to authenticate with their own credentials.

• Transient tracker store connection errors (e.g. Postgres restart) no longer flood logs with full stack traces on each retry attempt; the error message is preserved on one line.

• Fix resume_flow() marking the wrong AgentStackFrame as resumed when a flow contains multiple sub-agent frames, which prevented run_agent() from taking its resume branch.

• Upgrading from 3.15.x to 3.16.x no longer causes every subsequent message to fail with InvalidFlowStepIdException when a tracker was mid-flow at upgrade time and a flow step ID was renamed or removed. The flow executor now detects stale step IDs, logs a structured warning, removes the stale frame, and continues processing — rather than propagating the exception as a GraphComponentException.

[3.16.12] - 2026-06-02

Rasa Pro 3.16.12 (2026-06-02)

Bugfixes

• Fixed a performance regression in SQLTrackerStore.retrieve() introduced in v3.16, where every retrieve loaded the full conversation history before replay-safe session slicing. Retrieve now fetches a minimum event prefix by trying the most recent session boundaries first (up to three). When reconstruction fails with a dialogue-stack patch error, the store retries from the last stack event in that prefix whose patch applies to an empty stack. If all 3 attempts fail, the store falls back to loading the full conversation history as before. This greatly reduces database I/O and memory use for long-lived conversations while preserving replay-safe behaviour.
• Fixed graph schema validation on Python 3.13+ when components use PEP 604 union type hints (e.g. Domain | None), which previously raised a false GraphSchemaValidationException.
• Fixed MCP flow tool tracing metrics when _execute_mcp_tool_call mutates the shared initial_events list in place by capturing the pre-mutation event count for output slicing, and by detecting failures from InternalErrorPatternFlowStackFrame on the dialogue stack rather than from unchanged event list length.
• Updated idna, pip and urllib3 to address security vulnerabilities.

[3.16.11] - 2026-06-01

Rasa Pro 3.16.11 (2026-06-01)

Bugfixes

• Fixed DIETClassifier training failures on GPU with TensorFlow 2.19 and Keras 3, where training could fail with an InvalidArgumentError about unsupported SparseReshape operations on XLA GPU JIT devices. RasaModel now inherits from tensorflow.keras.Model instead of standalone Keras 3's Model, so it uses the same Keras stack as the rest of Rasa's TensorFlow code and respects TF_USE_LEGACY_KERAS=1.

[3.16.10] - 2026-05-22

Rasa Pro 3.16.10 (2026-05-22)

Bugfixes

• Fixed a KeyError in ContinueInterruptedPatternFlowStackFrame.from_dict that occurred when loading tracker events stored before the multi-interrupt fields were introduced. This caused repeated tracker store retries that blocked the server event loop, leading to liveness probe failures and pod restarts.

[3.16.9] - 2026-05-20

Rasa Pro 3.16.9 (2026-05-20)

Bugfixes

• Fix an interrupted sub-agent not being properly resumed after the user answers "Yes" to pattern_continue_interrupted. This is fixed by introducing a transient RESUMING agent state that sits between INTERRUPTED and WAITING_FOR_INPUT: resume_flow() now sets the agent frame to RESUMING instead of directly to WAITING_FOR_INPUT; run_agent() detects this and re-invokes the agent with resumed_after_interruption=True.

[3.16.8] - 2026-05-19

Rasa Pro 3.16.8 (2026-05-19)

Bugfixes

• Fixed InvalidFlowIdException and repeated error logs that occurred when a conversation tracker contained historical events referencing flows or slots that were subsequently removed from the model.

[3.16.7] - 2026-05-15

Rasa Pro 3.16.7 (2026-05-15)

Bugfixes

• Updated security-flagged dependencies:
  • litellm: 1.82.4 → 1.84.0
  • langchain: ~0.3.27 → ~1.2.16
  • langchain-community: ~0.3.29 → ~0.4.1
  • langchain-core: ~0.3.81 → ~1.3.2
  • langsmith: ~0.6.3 → ~0.7.31
  • langchain-qdrant: ~0.2.1 → ~1.1.0
  • langchain-openai: transitive → ~1.2.1 (new direct pin)
  • langchain-text-splitters: transitive → ~1.1.2 (new direct pin)
• The no_text_normalization parameter is now correctly passed to the Rime TTS WebSocket connection, allowing users to skip text normalization to reduce latency.

[3.16.6] - 2026-05-12

Rasa Pro 3.16.6 (2026-05-12)

Bugfixes

• Fix bug when user's message during bot's speech is queued after bot is done speaking. When interruptions are DISABLED, if the bot is speaking and the user talks over it, the bot should ignore that speech entirely. When interruptions are ENABLED, the bot should only stop and listen if the user says enough words to qualify as an interruption (based on the min_words setting).

• When interruptions are ENABLED, user message should be queued for processing only if it passes interruption criteria.

  If interruptions are disabled, user message should be queued only if user speaks during collect step utterance.

  Updated signature of following methods:

  • send_start_marker: send_start_marker(self, recipient_id: str) => send_start_marker(self, marker_input: MarkerInput)
  • send_intermediate_marker: send_intermediate_marker(self, recipient_id: str) => send_intermediate_marker(self, marker_input: MarkerInput)
  • send_end_marker: send_end_marker(self, recipient_id: str) => send_end_marker(self, marker_input: MarkerInput)
  • send_marker_message: send_marker_message(self, recipient_id: str) => send_marker_message(self, marker_input: MarkerInput)
  • create_marker_message: create_marker_message(self, recipient_id: str) => create_marker_message(self, marker_input: MarkerInput) -> MarkerMessageOutput

  where MarkerInput is a Pydantic-based class:

  class MarkerInput(BaseModel):
      recipient_id: str
      marker_type: MarkerType
      step_type: Optional[StepType] = None

• Remove the "hand over" command from the default GPT-5.1 prompt template. The command had already been removed from all other prompt templates in the 3.16.0 release.

• Fixed a bug where sub-agent flows would end immediately after interruption resumption instead of re-invoking the agent.

  When a flow containing a sub-agent call was interrupted (e.g., by a user digression) and then resumed via pattern_continue_interrupted, the agent frame's state remained INTERRUPTED. This caused the flow executor to skip the agent call and advance to END, triggering pattern_completed prematurely.

  The agent frame state is now correctly reset to WAITING_FOR_INPUT during flow resumption, ensuring the sub-agent is re-invoked with full conversation history as expected.

• Fixed MCP tracing attribute extraction for router-based model groups so tracing no longer crashes with KeyError('provider') when instrumenting ReAct sub-agent send_message calls.

• When interruptions are enabled and bot is not speaking, user utterances should not be checked if they are passing min_words threshold.

[3.16.5] - 2026-04-24

Rasa Pro 3.16.5 (2026-04-24)

Bugfixes

• Fixed a bug where the validator was mutating the allowed values list of categorical slots in-place, causing None entries to accumulate in slots used across multiple flows via subflows. This corrupted the flow retrieval FAISS embeddings during training.
• Fixed multi-agent flows looping back to an agent call when the top stack frame was interrupted instead of waiting for input, which could leave multiple agents active and fail graph execution. The loop-back shortcut now applies only when the agent frame is waiting for user input.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.16.4] - 2026-04-17

Rasa Pro 3.16.4 (2026-04-17)

Bugfixes

• Fixed tracing span attributes for LLM components that use model groups. Spans now carry the correct model group ID and, for router groups, the representative model's attributes (preferring OpenAI when present for prompt token counting). The llm_is_router_group attribute is also set on spans when a router group is active.
• Update cryptography to version 46.0.7 to fix a security vulnerability. For more details, see CVE-2026-39892.

[3.16.3] - 2026-04-09

Rasa Pro 3.16.3 (2026-04-09)

Improvements

• Replaced the generic "Start the server from your IDE" message after rasa tools init with IDE-specific instructions for Claude Code, Cursor, VS Code, and JetBrains.
• Added rasa tools status command that displays current project configuration and runtime readiness.

Bugfixes

• Fixed validation incorrectly rejecting custom ContextualResponseRephraser subclasses configured as nlg.type in endpoints.yml. The validator now resolves the class and checks inheritance instead of only accepting the literal string "rephrase".
• Updated PyJWT, werkzeug, aiohttp, orjson, pyasn1, requests, and ujsonto resolve security vulnerabilities.

[3.16.2] - 2026-04-02

Rasa Pro 3.16.2 (2026-04-02)

Bugfixes

• Fixed correction reset when a user answers the active collect slot and corrects another slot in the same turn.

  CorrectSlotsCommand.create_correction_frame no longer chooses the reset step only from slots in the correction payload. It also considers the current collect-information step and picks the earlier collect step in flow order, so the stack resets to the question currently being asked instead of jumping ahead to a later slot’s collect step (which skipped validation for the new value on the active slot).

[3.16.1] - 2026-04-01

Rasa Pro 3.16.1 (2026-04-01)

Bugfixes

• Fixed three bugs in ConcurrentRedisLockStore that caused an IndexError: deque index out of range crash in the PII anonymization and deletion cron jobs under multi-replica deployments.

  • get_lock() now returns None when no ticket keys exist in Redis (all tickets had expired), instead of returning an empty ConcurrentTicketLock object that caused downstream callers to crash.
  • save_lock() now skips the Redis write when the ticket has already expired (TTL ≤ 0), preventing a ResponseError from Redis.
  • save_lock() previously passed the absolute epoch expiry timestamp as the Redis EX TTL, resulting in ticket keys that effectively never expired (~55 years). The TTL is now correctly computed as a relative duration in seconds.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.16.0] - 2026-03-26

Rasa Pro 3.16.0 (2026-03-26)

Breaking changes

Upgrading may require config edits, code updates, or retraining. All incompatible changes for this release are listed below in one place.

• Default language models used by built-in LLM features have changed. If you need the same behavior as before, pin the previous model names in endpoints.yml.

  Component	New default model
  LLMBasedCommandGenerator	gpt-5.1-2025-11-13
  CompactLLMCommandGenerator	gpt-5.1-2025-11-13
  SearchReadyLLMCommandGenerator	gpt-5.1-2025-11-13
  ContextualResponseRephraser	gpt-5.1-2025-11-13
  IntentlessPolicy	gpt-5-mini-2025-08-07
  MCPBaseAgent	gpt-5.1-2025-11-13
  EnterpriseSearchPolicy	gpt-5-mini-2025-08-07
  LLMBasedRouter	gpt-5-mini-2025-08-07
  ConversationRephraser	gpt-5-mini-2025-08-07

• Default prompts for command generation changed. If you already set a custom prompt_template in config.yml, that file is still used.

• Default prompts for ReAct agents changed (task vs general assistant behavior and task_completed guidance). If you override templates, compare your custom files against the latest templates in rasa/agents/templates/.

, #4851, #4867: Agent interface changed.

1. process_output was removed from the shared agent protocol. Use process_agent_output for A2AAgent implementations and process_tool_output for ReAct/MCP agent implementations:

   # A2A agents
   async def process_agent_output(self, output: AgentOutput) -> AgentOutput
   
   # ReAct/MCP agents
   async def process_tool_output(
       self,
       current_iteration_tool_results: Dict[str, AgentToolResult],
       cumulative_tool_results: Dict[str, AgentToolResult],
       output_channel: Optional[OutputChannel] = None,
   ) -> List[Event]

2. run signature changed with optional cooperative cancellation support:

   async def run(
       self,
       input: "AgentInput",
       output_channel: Optional[OutputChannel] = None,
       cancellation_token: Optional["CancellationToken"] = None,
   ) -> "AgentOutput"

3. Custom tool callable signature changed:

   • Before: Callable[[Dict[str, Any]], AgentToolResult]
   • After: Callable[[Dict[str, Any], AgentToolContext], AgentToolResult]
   • Import AgentToolContext via from rasa.agents.schemas import AgentToolContext.

Deprecations and Removals

• Remove HumanHandoff command from the codebase and default prompt templates.
• Remove generic process_output method from AgentProtocol. In ReAct / MCP agents, use process_tool_output for tool-result-based post-processing. In A2AAgent, use process_agent_output method.

Features

• Added optional reasoning_effort on LLM model configuration for providers and models that support it (for example OpenAI and Azure OpenAI reasoning-capable models). By default, Rasa tries to apply the lowest appropriate reasoning_effort for the resolved model name (using provider metadata where available). We recommend using the lowest supported value unless you explicitly need deeper reasoning. Any value you set explicitly is left unchanged. Models that do not support this parameter do not receive a default. For Azure deployments configured without a concrete model name, Rasa can resolve the backing model and set reasoning_effort after the first probe response.

  Example configuration in endpoints.yml:

  model_groups:
    - id: openai-reasoning
      models:
        - provider: openai
          model: gpt-5-mini-2025-08-07
          reasoning_effort: "minimal"

• Added max_polling_time and polling_initial_delay to A2A agent configuration so polling parameters can be set per agent. Defaults remain 60 seconds and 0.5 seconds respectively when not set.

  To override polling for an A2A agent, set the options in the agent's configuration:

  agent:
    name: my_agent
    protocol: A2A
    # ...
  
  configuration:
    agent_card: "https://example.com/agent-card.json"
    max_polling_time: 120        # max total polling time in seconds (default: 60)
    polling_initial_delay: 1.0   # initial delay before first poll in seconds (default: 0.5)
    # ... other configuration (timeout, max_retries, etc.)

• Added filler message support for ReAct agents via a new enable_filler_messages config flag. Filler messages are enabled by default and are streamed to the user before tool execution, with streaming capability determined by a new supports_streaming property on OutputChannel.

  To disable filler messages for an agent, set enable_filler_messages: false in the agent's configuration:

  agent:
    name: my_agent
    # ...
  
  configuration:
    enable_filler_messages: false
    # ... other configuration (llm, timeout, etc.)

• Implements user-scoped tracker querying and durable conversation metadata across all tracker stores. Persists user_id and conversation_started_timestamp in the tracker object to use for querying and ordering:

  • serialisation now omits null user_id
  • deserialisation handles JSONDecodeError
  • Ensures conversation_started_timestamp backfilled on save/update for backward compatibility

  Summary of tracker store changes for user-scoped retrieval of conversation trackers:

  • SQL: introduces users table (sender_id→user_id, timestamp) with upsert per dialect; JOIN-based retrieval and DB-level ordering/pagination; cleanup on delete
  • Redis: maintains secondary index user_trackers:\{user_id\} (Sorted Set) for O(1) lookup; batch mget, robust deserialisation, index cleanup
  • Mongo: creates indices on user_id and (conversation_started_timestamp, sender_id); aggregation pipeline for ordering/pagination; restores user_id
  • Dynamo: saves user_id and conversation_started_timestamp (Decimal) in update_item; GSI support (user_id-index with sort on conversation_started_timestamp) with full-scan fallback

• Added optional user_id parameter to DialogueStateTracker to enable associating multiple conversations with a single end user.

• Implemented conversation lifecycle management with ConversationInactive and enhanced SessionEnded events. Conversations can now be marked as inactive (resumable via user message or ConversationResumed event) or terminated (permanently ended, blocking all further updates).

  Added a new session_config option start_session_after_expiry that controls what happens when a session expires (based on session_expiration_time).

  • When true (default), Rasa emits a SessionStarted event and starts a new session, ensuring backward compatibility.
  • When false, expired sessions simply continue without automatically starting a new session.

• Add REST API endpoint /users/{user_id}/trackers to retrieve all conversations for a given user.

  This admin endpoint provides user-scoped tracker querying with:

  • Authentication: Requires token or JWT authentication
  • Pagination: Supports limit and offset query parameters for efficient pagination
  • Event verbosity: Configurable via include_events query parameter (NONE, APPLIED, AFTER_RESTART, ALL)
  • Error handling: Proper validation and HTTP status codes for invalid inputs
  • Performance: Database-level sorting and pagination for efficient queries with 1000+ conversations

  Important: This is an admin endpoint that should be used as a proxy or in a backend service to manage what data is exposed to the client. Direct client access should be avoided to maintain security and control over sensitive user data.

  Example usage:

  GET /users/{user_id}/trackers?limit=50&offset=0

  Response format:

  {
    "conversations": [...],
    "limit": 50,
    "offset": 0
  }

• Implemented Multilingual Voice AI Agents. The configuration of Automatic Speech Recognition (ASR) and Text to Speech (TTS) changes through the conversation based on the value of built-in language slot. For each ASR and TTS Engine, users can configure the language, model or voice to be used for each value of language slot. In case of ASR Engines, reconfiguration requires a websocket reconnect. Rasa manages that on the fly to ensure that no speech is lost. The signature of certain methods in classes ASR_Engine and TTS_Engine have also changed. These methods are, init and from_config_dict.

• Implemented session-scoped conversations. Each conversation session now has a unique session_id (UUID v4) that is automatically generated and attached to all events within that session. This enables better tracking and analytics of user sessions. Session IDs are automatically generated on session start or resume events and propagated to event metadata.

  Added current_session_id field to DialogueStateTracker. The session ID is automatically:

  • Generated as a UUID4 when a new session starts or resumes. Generation triggers include:
    • Brand new conversations (first event on a tracker)
    • SlotSet for session metadata slot
    • action_session_start or SessionStarted events
    • ConversationResumed after inactivity
    • UserUttered reactivating an inactive conversation

  Added session_id to event metadata. The generated ID is automatically included in the metadata of all events created during a conversation session.

• Add timer_store configuration to endpoints.yml to enable configurable background session timer storage.

• Add MCP _meta parameter support so you can pass user-specific data to MCP servers without exposing it to the LLM.

  • Configuration: Optional meta_map per MCP server in endpoints.yml with:
    • static: Fixed key-value pairs always sent in _meta
    • from_slots: Slot-to-mcp key mappings; slot values are sent under the given param in _meta
  • Execution: Both flow-based and agent-based MCP tool calls build and send _meta when meta_map is configured. A compatibility layer supports current MCP SDK versions that do not expose meta on call_tool.
  • Agent-based calls: For from_slots, values are read from agent input only when the slot is present; slots that have been removed from agent input will not appear in _meta.

  Example endpoints.yml:

  mcp_servers:
    - name: internal_api_server
      url: http://internal-api:8000/mcp/
      type: http
      meta_map:
        static:
          api_version: "v2"
          source: "rasa_agent"
        from_slots:
          - slot: user_id
            param: user_id
          - slot: role
            param: user_role

  This replaces the need to override private Rasa methods (e.g. execute_mcp_tool, send_message) to inject auth or context into MCP requests.

• Implemented Session Timer System for automatic session expiration. Conversations are now automatically marked as inactive after a configurable timeout period, supporting both single-instance and horizontally-scaled deployments.

  • Timers are scheduled when a session starts and reset on each user message
  • When a timer expires, a ConversationInactive event is emitted
  • Race condition handling ensures correct behavior in multi-pod deployments
  • Redis-backed storage enables timer persistence across restarts
  • Automatic recovery of expired timers on server startup
  • Fallback to in-memory storage during Redis outages
  • SessionTimerStore / SessionTimerManager abstract base classes
  • InMemorySessionTimerStore / InMemorySessionTimerManager for single-instance deployments
  • RedisSessionTimerStore / RedisSessionTimerManager for distributed deployments

• Implemented out-of-the-box collection of customer satisfaction.

  • Added built-in pattern_customer_satisfaction for CSAT collection.
  • The pattern is triggered automatically via a link from pattern_completed when users decline to continue the conversation.
  • It collects a csat_score slot (categorical: satisfied/unsatisfied) and responds with appropriate thank-you messages.
  • The pattern, its responses (utter_ask_csat_score, utter_csat_thank_you_satisfied and utter_csat_thank_you_unsatisfied) and when the pattern is triggered are all fully customizable.

• Handle SessionEnded events returned from custom actions gracefully by stopping the prediction loop and flow executor when the conversation is terminated.

• Added the system action action_handoff_metric to mark conversations as "not contained" for containment metrics. Any conversation that includes this action emits an ActionExecuted Kafka event and is counted as not contained. The action is a no-op; no user-visible behavior or migration is required.

  Updated the default pattern_human_handoff flow to run action_handoff_metric before utter_human_handoff_not_available, so existing assistants automatically emit the metric when the human handoff pattern is used. If you have overridden pattern_human_handoff in your flows, you can add a step action: action_handoff_metric at the start of that flow to include those conversations in containment metrics.

• Add support for multiple Audio Formats in Rasa. These are G.711 μ-law (legacy) along with Linear PCM 24kHz and Linear PCM 48kHz.

• Extended support for Audio Formats, Updated Azure and Deepgram Automatic Speech Recognition (ASR) to support multiple audio formats. G.711 μ-law, Linear PCM 24kHz and Linear 48kHz. Updated Azure, Deepgram, and Cartesia Text to Speech (TTS) to support the same audio formats. Rime TTS only supports G.711 μ-law and Linear PCM 24kHz. Updated Jambonz and Audiocodes Stream Channel to use Linear PCM 24kHz audio when sending or receiving audio from the platform.

• Add prompt templates and the corresponding model mapping for GPT-5.1 and GPT-5.2.

• Update default models for LLM-based components: CompactLLMCommandGenerator, SearchReadyLLMCommandGenerator, LLMBasedCommandGenerator, ContextualResponseRephraser and MCPBaseAgent will use gpt-5.1-2025-11-13; EnterpriseSearchPolicy, LLMBasedRouter, ConversationRephraser and IntentlessPolicy will use gpt-5-mini-2025-08-07.

• Browser Audio channel uses Linear PCM 48kHz audio with ASR and TTS components. The sample rate of this encoding can be configured with sample_rate property of the channel. Supported values in kHz are 8000, 24000 and 48000 Voice Inspector has been updated to receive and play Linear PCM at any of the supported sample rates. The sample rate is sent to the Voice Inspector in a handshake message.

• ReAct agent now uses the streaming LLM API and consumes the response incrementally instead of buffering the full stream. Content tokens are delivered to the user in real time when the channel supports streaming. Tool calls are executed after the stream segment completes and their results are sent in one shot. This reduces perceived latency.

• Voice channels enforce two type of minimum gaps between bot audio: a shorter gap for regular back-to-back utterances, and a longer gap following filler messages (e.g., when streaming assistant text alongside tool calls from ReAct-style agents). After each message is sent, when the next message arrives, the channel checks the time since the last message was sent against the configured minimum delay. If more time is needed, generates silence audio for that duration, sends the silence to the stream, and then sends the next message. Consecutive turns are therefore naturally spaced.

• Add prompt templates and the corresponding model mapping for Claude Sonnet 4.5 and Claude Sonnet 4.6.

• Add public method process_tool_output() in MCPBaseAgent for implementing custom post-processing logic based on the tool results in each iteration.

• Support metadata pass-through in A2AAgent so that specialized data can be passed to the agent's backend via AgentInput.metadata without being processed by the LLM. The metadata is forwarded to the Message sent to the A2A server.

• • Add graceful cancellation of long-running A2A agent operations (both polling and streaming). When ConversationInactive event is added to the tracker after session_expiration_time minutes while an A2A task is in progress, the operation is now cancelled promptly instead of running until timeout. Note: This is a model breaking change. Please retrain your model.
  • Introduce a public method Agent.cancel_background_tasks(sender_id) that can be used to explicitly trigger the cancellation of the background A2A processing (both streaming and polling) from other components (e.g. from custom channels).
  • Add a new POST /cancel_background_tasks/<sender_id> endpoint in the REST channel that allows frontends and external systems to explicitly cancel background processing for a conversation.
  • Closing the Inspector browser tab now automatically cancels any in-flight A2A operations for that conversation.

• AgentInput.metadata is now propagated to custom tool execution. Tool executors receive (args, context), where args contains the LLM-generated tool parameters and context is an AgentToolContext. The request metadata is available via context.metadata and it can be set by the agent in the process_input method using the AgentInput.metadata attribute.

Improvements

• The continue-interrupted pattern is now triggered after a knowledge (e.g. EnterpriseSearch) answer, so users are asked if they want to resume previously interrupted flows when returning from a search.

• When an agent was interrupted and then resumed, it now gets reinvoked with the current context and a system instruction that it was interrupted and may use the new context or ask again. Previously, the agent only repeated its last message and did not run again.

• Add SessionPaused event which purpose is to be triggered by external systems when a session is paused. It has the following properties:

  • event - which is always set to session_paused
  • reason - which describes why the session was paused
  • timestamp - (optional) Unix timestamp, which indicates when the session was paused

  If timestamp is not provided, the current Unix timestamp of the Rasa Pro server will be used.

  This event can be used to notify pipelines bout the pause state of a session, allowing them to take appropriate actions based on the session's status.

  To use this event, external systems should trigger it when a session is paused, via POST conversations/<sender id>/tracker/events.

  Example of the payload for the POST request:

  {"event": "session_paused", "timestamp": 1761232628.512342, "reason": "Widget closed"}

• Refactor Copilot response handling to support structured streaming from the agent SDK copilot and improve text content processing. The new implementation properly segments text content parts into start, delta, and end events, handles controlled predictions, and fixes buffer handling issues in the legacy copilot handler.

• Events streamed to the event broker now include user_id in the message payload alongside sender_id. This allows downstream consumers to identify the end user associated with each conversation event.

• Add support to stream audio chunks from Azure Speech Service via websocket. Rasa will use streaming via websocket when response does not contain any SSML tags. If response contains any SSML tag, Rasa will default to use HTTPS to create audio chunks from Azure Speech Services. Improve interruption handling by:

  • stopping stream from Azure Speech Services when interruption happens
  • stopping to send audio chunks from Rasa to external voice service when interruption happens

• Improve action_session_start execution to ensure it runs only once (coordinating calls from the MessageProcessor and the FlowPolicy when it triggers pattern_session_start).

• Update eligibility and session-splitting logic for privacy cron jobs in line with the new session management improvements.

  • when USER_CHAT_INACTIVITY_IN_MINUTES is unset, “new” trackers are processed based on ConversationInactive/SessionEnded events and session_id grouping (with a grace period of min_after_session_end), while “legacy” trackers without session_id fall back to a fixed 30min + min_after_session_end threshold.
  • when the env var is set, time-based behavior remains but is now deprecated and applied per session.

  SQL tracker store: update() now supports a content-only path when the timestamp-based delete removes no rows (e.g. anonymization: same timestamps, content changed). In that case, the store either updates existing event rows in place where content differs or performs a full replace if event counts differ, so anonymized content is persisted correctly in a single atomic transaction.

  Anonymization cron job: the privacy manager now uses update(updated_tracker) for anonymization instead of delete-then-save. SQL tracker store behavior is aligned with MongoDB, Redis and DynamoDB (overwrite semantics), and anonymization with SQL completes in one atomic operation without a separate delete/save sequence.

• Cancel timer unconditionally when a SessionEnded event is appended via the REST API, regardless of whether execute_side_effects is requested.

• Add model_name to event metadata.

• Add tool_timeout for MCP/ReAct agents configuration to configure tool-call timeout in seconds.

• Enabled deletion of a Studio assistant during upload. Introduced a new --dangerously-delete-existing flag to rasa studio upload. When set, any existing Studio assistant with the same name is deleted before the upload proceeds, with no interactive confirmation.

• Added logic for signaling interruptions in tts streaming. Sendind a variation of clear command to stop streaming audio and clear the buffer to prepare for the next response.

• Added a warning log when carry_over_slots_to_new_session and start_session_after_expiry are both set to false in the domain session config, as carry_over_slots_to_new_session has no effect when no session boundary is created on expiry.

• Use the prompt template for gpt-5.1-2025-11-13 as new default prompt template in the CompactLLMCommandGenerator and SearchReadyLLMCommandGenerator.

• SQL and Mongo tracker stores now slice the latest conversation session using the same boundary as other tracker stores: ActionExecuted(action_session_start). That action is always executed by the message processor when a new session starts, while SessionStarted may be omitted by a custom action-session-start action. Incremental save offsets and SQL partial-fetch queries were updated to match this boundary.

• Add normalization of AgentOutput.events so entries from ReAct/A2A customizations may be either Rasa Event instances or serialized event dicts; dicts are parsed into proper events before flow handling (e.g. stack metadata attachment), and invalid entries are dropped with structured error logs.

• Declared safetensors and keras as optional NLU extras (pip install 'rasa-pro[nlu]' or full), so minimal installs no longer pull them in directly. regex is also declared optional and listed under those extras for consistency with WhitespaceTokenizer, but it remains installed on most minimal installs because the base dependency tiktoken (used by the LLM stack) depends on regex.

  The affected code paths lazy-import safetensors and regex; if either package is not installed, they raise MissingDependencyException with install instructions. keras is only pulled in for the TensorFlow model stack (rasa.utils.tensorflow.models), so it is optional for the same minimal-install story as safetensors.

• Improved sub-agent call steps that use exit_if: when the same ReAct agent runs again in one conversation, slots named in exit_if are cleared so a new run does not reuse values from the last finished run. Ongoing multi-turn agent sessions (waiting for the next user message) are unchanged.

• Two improvements to DynamoTrackerStore:

  • Idempotent save(): concurrent or retried saves no longer produce duplicate events. New conversations are written via put_item with attribute_not_exists(sender_id), ensuring the full tracker is stored exactly once. Subsequent saves append only new last-turn events via a conditional update_item guarded by last_event_timestamp, which atomically rejects any append whose events are already stored — without requiring a read-before-write.
  • Reused boto3 resource: boto3.resource("dynamodb") is now instantiated once in __init__ as self._dynamo and shared across all methods, eliminating repeated resource construction per call. The connection pool size is configurable via the max_pool_connections constructor argument (default: 50) to prevent urllib3 pool saturation under concurrent load.

• The GET /users/{user_id}/trackers endpoint no longer replays conversation history through DialogueStateTracker.from_dict() and current_state() on every request. Each tracker store now implements get_serialized_trackers_by_user_id(), which returns serialized event dicts directly from storage without event replay.

  Additional store-specific improvements included with this change:

  • SQL: replaced an N+1 query loop (one retrieve_full_tracker call per conversation) with two bulk queries — one paginated users table query and one events IN (...) fetch — eliminating query count growth with conversation volume.
  • Redis: orphaned sorted-set index entries are now removed in a single batched ZREM call instead of one call per orphan.
  • All stores: current_session_id is now derived consistently from the last event's metadata rather than from stored state, correctly returning None when the last event is ConversationInactive.

Bugfixes

• Restarted task-oriented ReAct agents no longer exit immediately. When an agent is restarted, the previous run's user/assistant messages are now marked in the conversation so the model does not set slots from them; the system prompt instructs the agent to treat the current interaction as a fresh start for slot collection.

• Fix potential Tensor shape mismatch error in TEDPolicy and DIETClassifier.

• Fix bug with the missing language_data module by installing langcodes with the data extra dependency.

• Fixed language parameter in RimeTTS configuration to be passed correctly to the Rime API. Temporarily changed RimeTTS to non-streaming input mode as a workaround for a Rime API bug that impacts conversation experience; audio is now generated from the complete response at once.

• Update mcp version to ~1.23.0 to address security vulnerability CVE-2025-66416.

• Fix bug in CommandPayloadReader where regex matching did not account for list slots, leading to incorrect parsing of slot keys and values. Now, slot names and values are correctly extracted even if a list is provided.

• Previously, DialogueStateTracker.has_coexistence_routing_slot could incorrectly return True when the tracker was created without domain slots (i.e., using AnySlotDict), because AnySlotDict pretends all slots exist. Now, the property returns False in that case, so the routing slot is only considered present if it is actually defined in the domain.

• Fix LLM prompt template loading to raise an error instead of just printing a warning when a custom prompt file is missing or cannot be read.

• Fix AttributeError: 'str' object has no attribute 'pop' when using KnowledgeAnswerCommand with tracing enabled.

• Fix UnboundLocalError in E2ETestRunner._handle_fail_diff that caused e2e tests to crash when processing slot_was_not_set assertion failures.

• Fixed button payloads with nested parentheses (e.g., /SetSlots(slot=value)) not being parsed correctly in rasa shell. The payload was incorrectly stripped of its /SetSlots prefix, causing it to be treated as text instead of a slot-setting command.

• Add missing deprecation warning for legacy RASA_PRO_LICENSE environment variable. Update license validation error messages to reference the actual environment variable used.

• Include response button titles in conversation history in prompts of LLM based components.

• Fixed e2e test coverage report to correctly track coverage per flow. Each flow now appears as a separate entry with accurate coverage percentages. Additionally, call/link steps and collect steps with prefilled slots are now properly marked as visited.

• Fix DUT fails with AttributeError when running with custom CommandGenerator

• Fix OAuth2AuthStrategy to use URL-encoded parameters with Basic Authentication to fetch access token

• Remove config file content from endpoint read success logs to prevent sensitive data exposure.

• The MCP task agent now includes slot changes in the agent output when the state is INPUT_REQUIRED or when max iterations is reached. Previously, slots set via set_slot tools were updated in memory but not forwarded in the output, so they were not persisted until exit conditions were met.

• Fix race conditions in PII cron jobs by requiring a LockStore for BackgroundPrivacyManager and acquiring a per-sender_id lock while anonymization/deletion jobs read-modify-write trackers.

• Rasa needs to check if it is in listening mode before it triggers the silence timeout watcher. Silence timeout watcher will be triggered if all are true:

  • User is not speaking
  • Bot is not speaking
  • Rasa is in listening mode

  If any of the conditions above are not true, Rasa will cancel the silence timeout watcher.

• Channel specific silence_timeout can now be set in credentials file. Fixes the type error. If silence_timeout is not set the channel will use global value.

• Stop silence timeout immediately when interruption is detected.

• When an A2A task has status input_required, Rasa now populates AgentOutput.structured_results from both task.artifacts and DataParts in task.status.message.parts. Previously only the text message was returned and structured data was omitted, so clients can now process artifact data when the agent is waiting for user input.

• Fix PII redaction of user or bot messages when slot values do not match the original message text. This can occur when the original messages are multi-line or TTS-style bot read-backs.

• Update setuptools to 80.10.2 to address vulnerability https://osv.dev/vulnerability/DEBIAN-CVE-2026-23949. Replaced randomname with duoname (no dependencies) library.

• Fixed the backend tracing wrapper (e.g., Jaeger or OTLP) around send_message to correctly forward the output_channel argument to the underlying implementation.

• When an agent or MCP tool call returns a fatal error (e.g. internal server error), the user no longer receives two bot messages. Previously both the internal error message ("Sorry, I am having trouble with that...") and the flow-cancelled message ("Okay, stopping...") were shown. Now only the internal error message is shown; the cancel pattern is not pushed for system-initiated failures, so the flow is still marked ended for tracking but the confusing cancel utterance is omitted.

• Add rephrase endpoint validation by treating defaults-only misconfiguration as warning and user-defined rephrase misconfiguration as error.

• Issue a warning when ASR and TTS language_map is missing languages from additional_language property read from config.yml.

• Fixed replay-safe latest-session handling for all built-in tracker stores. Tracker store retrieve method uses get_latest_replay_safe_session_tracker, widening the replay prefix when stack patches cannot be applied on the latest-session slice alone.

• Fixed default_language raising RasaException("No default language configured") even when language: was properly set in config.yml. It happened when the langauge slot was reset (set to None).

• Improved OpenTelemetry histograms for MCP agent LLM usage: mcp_agent_llm_prompt_token_usage and mcp_agent_llm_response_duration now record a consistent, call-scoped set of attributes (agent_name, execution_context, and protocol_type), making metrics easier to aggregate.

• Fixes Implemented:

  1. Both default and customized action_session_start now raise a Rasa exception when the session is already started for the current message. This exception is caught and handled by MessageProcessor which appends ActionExecutionRejected event and continues processing the message. This ensures that the message processing flow is not interrupted by the exception and allows for proper handling of the session start action. This change was required to avoid generating new session ids when this action is executed multiple times in the same time-bound tracker session.

  2. Fix conversation stalling when a flow contains an explicit action_session_start step and the session is already started for the current message.

  • The root cause explanation: The attempt to execute action_session_start a 2nd time for an existing session is rejected. The action execution rejection propagates to the DefaultPolicyPredictionEnsemble, which zeroes out action_session_start's confidence across all policy predictions. Since FlowPolicy only predicted that one action (with confidence 1.0), zeroing it leaves only the default action_listen at 0.0.
  • The fix: detect this condition in FlowPolicy and skip the step immediately, letting FlowPolicy continue the loop and return the actual next flow step's action.

• Fixed empty assistant bubbles in Socket.IO channels (including Inspector) when the model returned whitespace-only or paragraph-only text. Non-streaming LLM responses are now stripped before send and blank segments from \n\n splitting are no longer emitted.

• Fixed three bugs in DynamoTrackerStore:

  • JsonPatchConflict on tracker retrieval: duplicate events appended by concurrent or retried save() calls caused jsonpatch to attempt operations on already-mutated dialogue stack fields (e.g. removing frame_type a second time), raising an unhandled exception. Duplicate events are now removed on load via _deduplicate_events(), which deduplicates by full event content to safely handle multiple legitimate events sharing the same timestamp.
  • Event accumulation across trackers in get_trackers_by_user_id: events_with_floats was declared outside the per-item loop in _process_dynamodb_items, causing each reconstructed tracker to accumulate all preceding trackers' events. Each tracker now gets its own isolated events list.
  • Pre-UserUttered events lost on first save: save() only appended last-turn events (from the latest UserUttered timestamp onwards), so events emitted before the first user message — such as action_session_start, session_started, and initial slot events — were never persisted for new conversations. First saves now use put_item to store the complete tracker state.

• Upgrade langsmith dependency to 0.6.3 to address CVE-2026-25528.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.15.29] - 2026-06-23

Rasa Pro 3.15.29 (2026-06-23)

Bugfixes

• Bumped langchain-classic (~1.0.7, fixes CVE-2026-45134) and starlette (~1.3.1, fixes CVE-2026-54283) on 3.15.x to resolve high-severity dependency security vulnerabilities. Also raised the langchain-core floor to ~1.3.3, which is required by langchain-classic 1.0.7.

[3.15.28] - 2026-06-22

Rasa Pro 3.15.28 (2026-06-22)

Improvements

• A2A agent validation now requires TICKET_LOCK_LIFETIME to be at least as long as max_polling_time, preventing the conversation lock from expiring while A2A polling is still in progress.
• Added configure_litellm_for_rasa() in a lightweight _litellm_config module to centralize LiteLLM setup: it suppresses print-based provider-list debug footers via litellm.suppress_debug_info and sets the LiteLLM logger to WARNING, reducing log noise during LLM, embedding, and router client initialization without pulling in AWS provider dependencies.

Bugfixes

• Fixed MCP task agents exposing set_slot_* tools and listing slots in the task prompt for controlled-only or NLU-only slots in exit_if. Added a slot_mappings_allow_llm_fill helper so tool exposure, prompt rendering, and should_slot_be_set use the same rule for whether the LLM may fill a slot from its mappings.

• Fixed rasa data validate incorrectly rejecting call steps that target subagents or MCP tools when no agent configuration is loaded, instead of emitting a warning.

• Fixed suppress_logs so that concurrent async calls no longer corrupt the process-wide log level, preventing logs from disappearing permanently until pod restart.

• Refresh the AWS ElastiCache IAM presigned URL before each Redis (re)connect when within 60 seconds of expiry, so lock store and timer store operations no longer fail after >30 minutes of pro-server idleness.

• Message processing no longer crashes with ValueError: Could not find a user flow frame to cancel when an action fails (for example action_session_start returning HTTP 413) on a session whose residual dialogue stack contains only pattern frames; the internal-error pattern is pushed without attempting to cancel a non-existent user flow.

• Upgrade langsmith dependency to 0.8.5 and starlette to 1.0.1 to address CVE-2026-45134 and CVE-2026-48710.

• Fixed a leaked privacy-manager background scheduler thread that kept rasa test e2e (and the finetuning and dialogue-understanding test commands) from exiting cleanly.

  The privacy manager's background thread is now stopped on every path that retires an agent — the test CLI commands, server model reload/unload, and builder agent swaps — instead of only on server shutdown.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.15.27] - 2026-06-04

Rasa Pro 3.15.27 (2026-06-04)

Bugfixes

• Fixed a spurious "Fatal error on SSL transport" traceback printed to stderr on Windows after rasa test e2e completes by explicitly closing litellm's shared httpx sessions before the event loop shuts down.

• Fixed a crash in rasa-tools where training an assistant with a sub-agent MCP server configured caused a RuntimeError: Attempted to exit cancel scope in a different task than it was entered in, crashing the stdio server and making the next tool call fail with MCP error -32000: Connection closed.

• Fix two bugs causing random crashes and resource leaks when running Rasa with --workers 2+: a pre-fork event loop left open during JWT setup, and litellm async HTTP clients not being closed on worker shutdown.

  Root cause — Sanic uses fork-based multi-worker mode (legacy=True). create_app() was calling asyncio.new_event_loop() + set_event_loop() in the main process to satisfy sanic-jwt's Initialize(), then leaving the loop open. Forked workers inherited its kqueue/epoll file descriptors, conflicting with each worker's own loop (~50% crash at startup). Separately, close_resources() never called litellm.close_litellm_async_clients(), leaving cached async HTTP clients open when workers exit.

• An explicitly configured api_key in the rasa provider model config now takes precedence over the Rasa license, allowing custom endpoints to authenticate with their own credentials.

• Transient tracker store connection errors (e.g. Postgres restart) no longer flood logs with full stack traces on each retry attempt; the error message is preserved on one line.

• Fix resume_flow() marking the wrong AgentStackFrame as resumed when a flow contains multiple sub-agent frames, which prevented run_agent() from taking its resume branch.

• Fixed multiple Langfuse session IDs being created for a single Dialogue Understanding Test (DUT) run by using the original test-case sender ID as the session ID for all per-step LLM traces.

• Upgrading from 3.15.x to 3.16.x no longer causes every subsequent message to fail with InvalidFlowStepIdException when a tracker was mid-flow at upgrade time and a flow step ID was renamed or removed. The flow executor now detects stale step IDs, logs a structured warning, removes the stale frame, and continues processing — rather than propagating the exception as a GraphComponentException.

[3.15.26] - 2026-06-02

Rasa Pro 3.15.26 (2026-06-02)

Bugfixes

• Fixed graph schema validation on Python 3.13+ when components use PEP 604 union type hints (e.g. Domain | None), which previously raised a false GraphSchemaValidationException.
• Fixed MCP flow tool tracing metrics when _execute_mcp_tool_call mutates the shared initial_events list in place by capturing the pre-mutation event count for output slicing, and by detecting failures from InternalErrorPatternFlowStackFrame on the dialogue stack rather than from unchanged event list length.
• Updated idna, pip and urllib3 to address security vulnerabilities.

[3.15.25] - 2026-06-01

Rasa Pro 3.15.25 (2026-06-01)

Bugfixes

• Fixed DIETClassifier training failures on GPU with TensorFlow 2.19 and Keras 3, where training could fail with an InvalidArgumentError about unsupported SparseReshape operations on XLA GPU JIT devices. RasaModel now inherits from tensorflow.keras.Model instead of standalone Keras 3's Model, so it uses the same Keras stack as the rest of Rasa's TensorFlow code and respects TF_USE_LEGACY_KERAS=1.

[3.15.24] - 2026-05-22

Rasa Pro 3.15.24 (2026-05-22)

Bugfixes

• Fix an interrupted sub-agent not being properly resumed after the user answers "Yes" to pattern_continue_interrupted. This is fixed by introducing a transient RESUMING agent state that sits between INTERRUPTED and WAITING_FOR_INPUT: resume_flow() now sets the agent frame to RESUMING instead of directly to WAITING_FOR_INPUT; run_agent() detects this and dispatches to _handle_resume_interrupted_agent, which replays the agent's last message and transitions the frame back to WAITING_FOR_INPUT.
• Fixed a KeyError in ContinueInterruptedPatternFlowStackFrame.from_dict that occurred when loading tracker events stored before the multi-interrupt fields were introduced. This caused repeated tracker store retries that blocked the server event loop, leading to liveness probe failures and pod restarts.

[3.15.23] - 2026-05-15

Rasa Pro 3.15.23 (2026-05-15)

Bugfixes

• Updated security-flagged dependencies:
  • litellm: 1.82.4 → 1.84.0
  • langchain: ~0.3.27 → ~1.2.16
  • langchain-community: ~0.3.29 → ~0.4.1
  • langchain-core: ~0.3.81 → ~1.3.2
  • langsmith: ~0.6.3 → ~0.7.31
  • langchain-qdrant: ~0.2.1 → ~1.1.0
  • langchain-openai: transitive → ~1.2.1 (new direct pin)
  • langchain-text-splitters: transitive → ~1.1.2 (new direct pin)
• The no_text_normalization parameter is now correctly passed to the Rime TTS WebSocket connection, allowing users to skip text normalization to reduce latency.

[3.15.22] - 2026-05-06

Rasa Pro 3.15.22 (2026-05-06)

Bugfixes

• Fixed a bug where sub-agent flows would end immediately after interruption resumption instead of re-invoking the agent.

  When a flow containing a sub-agent call was interrupted (e.g., by a user digression) and then resumed via pattern_continue_interrupted, the agent frame's state remained INTERRUPTED. This caused the flow executor to skip the agent call and advance to END, triggering pattern_completed prematurely.

  The agent frame state is now correctly reset to WAITING_FOR_INPUT during flow resumption, ensuring the sub-agent is re-invoked with full conversation history as expected.

• Fixed MCP tracing attribute extraction for router-based model groups so tracing no longer crashes with KeyError('provider') when instrumenting ReAct sub-agent send_message calls.

• When interruptions are enabled and bot is not speaking, user utterances should not be checked if they are passing min_words threshold.

[3.15.21] - 2026-04-24

Rasa Pro 3.15.21 (2026-04-24)

Bugfixes

• Fixed a bug where the validator was mutating the allowed values list of categorical slots in-place, causing None entries to accumulate in slots used across multiple flows via subflows. This corrupted the flow retrieval FAISS embeddings during training.
• Fixed multi-agent flows looping back to an agent call when the top stack frame was interrupted instead of waiting for input, which could leave multiple agents active and fail graph execution. The loop-back shortcut now applies only when the agent frame is waiting for user input.
• Fixed tracing span attributes for LLM components that use model groups. Spans now carry the correct model group ID and, for router groups, the representative model's attributes (preferring OpenAI when present for prompt token counting). The llm_is_router_group attribute is also set on spans when a router group is active.

[3.15.20] - 2026-04-10

Rasa Pro 3.15.20 (2026-04-10)

Bugfixes

• When interruptions are ENABLED, user message should be queued for processing only if it passes interruption criteria.

• If interruptions are disabled, user message should be queued only if user speaks during collect step utterance.

  Updated signature of following methods:

  • send_start_marker: send_start_marker(self, recipient_id: str) => send_start_marker(self, marker_input: MarkerInput)
  • send_intermediate_marker: send_intermediate_marker(self, recipient_id: str) => send_intermediate_marker(self, marker_input: MarkerInput)
  • send_end_marker: send_end_marker(self, recipient_id: str) => send_end_marker(self, marker_input: MarkerInput)
  • send_marker_message: send_marker_message(self, recipient_id: str) => send_marker_message(self, marker_input: MarkerInput)
  • create_marker_message: create_marker_message(self, recipient_id: str) => create_marker_message(self, marker_input: MarkerInput) -> MarkerMessageOutput

  where MarkerInput is a Pydantic-based class:

  class MarkerInput(BaseModel):
      recipient_id: str
      marker_type: MarkerType
      step_type: Optional[StepType] = None

• Fixed validation incorrectly rejecting custom ContextualResponseRephraser subclasses configured as nlg.type in endpoints.yml. The validator now resolves the class and checks inheritance instead of only accepting the literal string "rephrase".

• Updated PyJWT, werkzeug, aiohttp, orjson, pyasn1, requests, and ujsonto resolve security vulnerabilities.

[3.15.19] - 2026-04-02

Rasa Pro 3.15.19 (2026-04-02)

Improvements

• Declared safetensors and keras as optional NLU extras (pip install 'rasa-pro[nlu]' or full), so minimal installs no longer pull them in directly. regex is also declared optional and listed under those extras for consistency with WhitespaceTokenizer, but it remains installed on most minimal installs because the base dependency tiktoken (used by the LLM stack) depends on regex.

  The affected code paths lazy-import safetensors and regex; if either package is not installed, they raise MissingDependencyException with install instructions. keras is only pulled in for the TensorFlow model stack (rasa.utils.tensorflow.models), so it is optional for the same minimal-install story as safetensors.

• Improved sub-agent call steps that use exit_if: when the same ReAct agent runs again in one conversation, slots named in exit_if are cleared so a new run does not reuse values from the last finished run. Ongoing multi-turn agent sessions (waiting for the next user message) are unchanged.

Bugfixes

• When an agent or MCP tool call returns a fatal error (e.g. internal server error), the user no longer receives two bot messages. Previously both the internal error message ("Sorry, I am having trouble with that...") and the flow-cancelled message ("Okay, stopping...") were shown. Now only the internal error message is shown; the cancel pattern is not pushed for system-initiated failures, so the flow is still marked ended for tracking but the confusing cancel utterance is omitted.

• Add rephrase endpoint validation by treating defaults-only misconfiguration as warning and user-defined rephrase misconfiguration as error.

• Improved OpenTelemetry histograms for MCP agent LLM usage: mcp_agent_llm_prompt_token_usage and mcp_agent_llm_response_duration now record a consistent, call-scoped set of attributes (agent_name, execution_context, and protocol_type), making metrics easier to aggregate.

• Upgrade langsmith dependency to 0.6.3 to address CVE-2026-25528.

• Fixed three bugs in ConcurrentRedisLockStore that caused an IndexError: deque index out of range crash in the PII anonymization and deletion cron jobs under multi-replica deployments.

  • get_lock() now returns None when no ticket keys exist in Redis (all tickets had expired), instead of returning an empty ConcurrentTicketLock object that caused downstream callers to crash.
  • save_lock() now skips the Redis write when the ticket has already expired (TTL ≤ 0), preventing a ResponseError from Redis.
  • save_lock() previously passed the absolute epoch expiry timestamp as the Redis EX TTL, resulting in ticket keys that effectively never expired (~55 years). The TTL is now correctly computed as a relative duration in seconds.

• Fixed correction reset when a user answers the active collect slot and corrects another slot in the same turn.

  CorrectSlotsCommand.create_correction_frame no longer chooses the reset step only from slots in the correction payload. It also considers the current collect-information step and picks the earlier collect step in flow order, so the stack resets to the question currently being asked instead of jumping ahead to a later slot’s collect step (which skipped validation for the new value on the active slot).

• When interruptions are ENABLED, user message should be queued for processing only if it passes interruption criteria.

[3.15.18] - 2026-03-16

Rasa Pro 3.15.18 (2026-03-16)

Bugfixes

• Restarted task-oriented ReAct agents no longer exit immediately. When an agent is restarted, the previous run's user/assistant messages are now marked in the conversation so the model does not set slots from them; the system prompt instructs the agent to treat the current interaction as a fresh start for slot collection.
• When an A2A task has status input_required, Rasa now populates AgentOutput.structured_results from both task.artifacts and DataParts in task.status.message.parts. Previously only the text message was returned and structured data was omitted, so clients can now process artifact data when the agent is waiting for user input.
• Fix PII redaction of user or bot messages when slot values do not match the original message text. This can occur when the original messages are multi-line or TTS-style bot read-backs.
• Update setuptools to 80.10.2 to address vulnerability https://osv.dev/vulnerability/DEBIAN-CVE-2026-23949. Replaced randomname with duoname (no dependencies) library.
• Fixed the backend tracing wrapper (e.g., Jaeger or OTLP) around send_message to correctly forward the output_channel argument to the underlying implementation.

[3.15.17] - 2026-03-11

Rasa Pro 3.15.17 (2026-03-11)

Deprecations and Removals

• Removed an explicit deprecation warning for the license varible RASA_PRO_LICENSE to provide clarity that we will continue to support both this and the newer RASA_LICENSE variable for the forseeable future.

Bugfixes

• Fixes MCP tool result handling to now process results that contain only structuredContent (and no or empty content). Previously, an empty content field caused the result to be skipped and slots were not set from structuredContent.

• Downgraded the A2A polling "waiting to poll again" log from ERROR to INFO so normal polling no longer triggers false alerts.

• SQL tracker store: update() now supports a content-only path when the timestamp-based delete removes no rows (e.g. anonymization: same timestamps, content changed). In that case, the store either updates existing event rows in place where content differs or performs a full replace if event counts differ, so anonymized content is persisted correctly in a single atomic transaction.

  Anonymization cron job: the privacy manager now uses update(updated_tracker) for anonymization instead of delete-then-save. SQL tracker store behavior is aligned with MongoDB, Redis and DynamoDB (overwrite semantics), and anonymization with SQL completes in one atomic operation without a separate delete/save sequence.

• Prevent JsonPatchConflict exceptions raised when tracker is split in sub-sessions during PII anonymization cron jobs. Replace usage of a utility function that was recreating tracker objects using sub-sessions with an approach retrieving list of events for each sub-session instead.

[3.15.16] - 2026-03-06

Rasa Pro 3.15.16 (2026-03-06)

Bugfixes

• When a flow had Agent A → flow steps → Agent B and the user restarted Agent A while Agent B was already started (then interrupted), execution after the restarted Agent A completed would jump back to Agent B and skip the flow steps between A and B. Flow steps between agents are now executed correctly, and the previously interrupted Agent B is resumed instead of started from scratch.
• Channel specific silence_timeout can now be set in credentials file. Fixes the type error. If silence_timeout is not set the channel will use global value.

[3.15.15] - 2026-03-03

Rasa Pro 3.15.15 (2026-03-03)

Bugfixes

• • Kafka event broker background polling thread caused test hangs and didn't allow the process to exit: The background poll thread is now a daemon thread, so the process can terminate when the broker is not closed explicitly (e.g. in tests or after an abrupt exit). The close() method uses a 5-second join timeout so it does not block forever if the poll thread is stuck (e.g. when the broker is unreachable).
  • Background polling: The poll loop now runs for the whole lifetime of the broker, even before the producer is created. When the producer is not ready, the loop sleeps briefly instead of exiting. This allows the IAM OAuth callback to be triggered when using SASL_SSL with AWS IAM (MSK), so tokens can be refreshed as the library calls poll().
  • Connection keepalive (optional): New broker options socket_keepalive_enable, reconnect_backoff_ms, reconnect_backoff_max_ms, and topic_metadata_refresh_interval_ms let you reduce idle connection drops. When socket_keepalive_enable is true, these are passed to the underlying producer.

• Fixes race conditions in PII cron jobs by requiring a LockStore for BackgroundPrivacyManager and acquiring a per-sender_id lock while anonymization/deletion jobs read-modify-write trackers.

  Implemented measures to defend against JSON patch issues and ensure correct event order.

  • Deletion cron job: When reconstructing the tracker from retained events after deleting old sessions, the deletion job now catches JsonPatchException and JsonPointerException (e.g. from invalid or inconsistent dialogue stack updates). On failure, the tracker is left unchanged and an error is logged instead of overwriting or deleting data, so no data loss occurs. Users can resolve stack-related issues by appending a Restarted event to the tracker if needed.
  • Anonymization cron job: Events from already-anonymized, processed, and ineligible sessions are now sorted by timestamp before rebuilding the tracker, with original index used as a tie-breaker when timestamps are identical. This keeps replay order chronologically consistent and ensures dialogue stack updates and other order-dependent events are applied in the correct sequence.

• ReAct agents now filter conversation events to user and bot utterances before building assistant and user role messages, and the default context limit is 10 utterance messages. Previously, the last 20 events were taken first and then filtered, so fewer user and assistant messages were included in the context.

[3.15.14] - 2026-02-26

Rasa Pro 3.15.14 (2026-02-26)

Bugfixes

• The MCP task agent now includes slot changes in the agent output when the state is INPUT_REQUIRED or when max iterations is reached. Previously, slots set via set_slot tools were updated in memory but not forwarded in the output, so they were not persisted until exit conditions were met.

[3.15.13] - 2026-02-25

Rasa Pro 3.15.13 (2026-02-25)

Bugfixes

• Upgraded litellm to 1.81.15 so that strict: True can be passed to Bedrock and tool calling has the same guarantees as with OpenAI. Upgraded openai to >=2.8.0 to satisfy litellm's dependency.
• Unquoted environment variable references for MCP and A2A OAuth credentials (client_id, client_secret) in endpoints.yml are now accepted. Previously, training failed unless these values were wrapped in double quotes (e.g. "${MCP_CLIENT_SECRET}").
• Move MCP task agent exit condition evaluation to after process_output, so that SlotSet events added by custom process_output implementations are considered when checking exit_if conditions.

[3.15.12] - 2026-02-23

Rasa Pro 3.15.12 (2026-02-23)

Bugfixes

• E2E fixture resolution and conftest hierarchy:

  Fixture resolution now uses a conftest-style hierarchy, with local overrides taking precedence over global fixtures. This is a change from the previous behavior where all fixtures were merged into a single list, which could lead to silent data loss if duplicate fixture names were used. Now, every test case has its own resolved set of fixtures, and duplicate fixture names are allowed when the intent is "override".

  Details:

  • Conftest-style hierarchy: Fixtures can be defined at root or folder level (e.g. conftest.yml / conftest.yaml); all e2e tests under that path see them.
  • Single-file run parity: Running one test file resolves fixtures the same way as when that file is run as part of the full suite (global/conftest fixtures are loaded for that file’s path).
  • Runtime namespace: Each test case has its own resolved set of fixtures; there is no shared mutable fixture state between tests. Duplicate fixture names in different files are allowed when the intent is “override” (no need to remove or rename for full-suite runs).
  • Override semantics: Local (file-level) fixtures override folder-level; folder-level overrides root. Within a single file, duplicate fixture names remain an error.

• Update cryptography dependency to 46.0.5 to address CVE-2026-26007. Update pillow dependency to 12.1.1 to address CVE-2026-25990. Update pyasn1 dependency to 0.6.2 to address CVE-2026-23490. Update python-multipart dependency to 0.0.22 to address CVE-2026-24486.

[3.15.11] - 2026-02-18

Rasa Pro 3.15.11 (2026-02-18)

Bugfixes

• Upgrade litellm to 1.80.0.
• Remove config file content from endpoint read success logs to prevent sensitive data exposure.
• Update protobuf to v5.29.6 to address CVE-2026-0994. Update wheel to v0.46.3 to address CVE-2026-24049.

[3.15.10] - 2026-02-10

Rasa Pro 3.15.10 (2026-02-10)

Improvements

• Flows can now link to pattern_search (e.g. to trigger RAG or branch on knowledge-based search).

Bugfixes

• Added interruption handling for final transcripts. Moved interruption check to eliminate interruption handling delay.
• Fix Enterprise Search Policy triggering utter_ask_rephrase instead of utter_no_relevant_answer_found when the vector store search returns no matching documents. The pattern_cannot_handle now correctly receives the cannot_handle_no_relevant_answer as a context.reason in all cases where no relevant answer is found, not only when the optional relevancy check rejects the answer.

[3.15.9] - 2026-02-05

Rasa Pro 3.15.9 (2026-02-05)

Bugfixes

• Upgrade Keras to 3.12.1 to address CVE-2026-0897.
• Fixed language parameter in RimeTTS configuration to be passed correctly to the Rime API. Temporarily changed RimeTTS to non-streaming input mode as a workaround for a Rime API bug that impacts conversation experience; audio is now generated from the complete response at once.

[3.15.8] - 2026-01-26

Rasa Pro 3.15.8 (2026-01-26)

Bugfixes

• Update azure-core to 1.38.0 to address CVE-2026-21226.

[3.15.7] - 2026-01-21

Rasa Pro 3.15.7 (2026-01-21)

Bugfixes

• Fixed e2e test coverage report to correctly track coverage per flow. Each flow now appears as a separate entry with accurate coverage percentages. Additionally, call/link steps and collect steps with prefilled slots are now properly marked as visited.
• Fix OAuth2AuthStrategy to use URL-encoded parameters with Basic Authentication to fetch access token

[3.15.6] - 2026-01-15

Rasa Pro 3.15.6 (2026-01-15)

Bugfixes

• Fixed an issue where storing documents in FAISS individually could hit the token limit. Now, documents are stored in batches to avoid exceeding the token limit.
• Throw error when duplicate fixtures are found in the same file or across files.
• SessionEnded does not reset the tracker anymore.
• Updated werkzeug, aiohttp, filelock, fonttools, marshmallow, urllib3 and langchain-coreto resolve security vulnerabilities.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.15.5] - 2025-12-29

Rasa Pro 3.15.5 (2025-12-29)

Bugfixes

• Include response button titles in conversation history in prompts of LLM based components.
• Fix OpenTelemetry exporter Invalid type <class 'NoneType'> of value None error when recording the request duration metric for requests made from Rasa server to the custom action server. This error occurred because the url parameter was not being passed to the method that records the request duration metric, resulting in a NoneType value.
• Fix DUT fails with AttributeError when running with custom CommandGenerator

[3.15.4] - 2025-12-19

Rasa Pro 3.15.4 (2025-12-19)

Bugfixes

• Fixed token expiration validation failing on servers running in non-UTC timezones. Token expiration checks now use timezone-aware UTC datetimes consistently, preventing premature "access token expired" errors on systems in timezones like UTC+8.

[3.15.3] - 2025-12-11

Rasa Pro 3.15.3 (2025-12-11)

Bugfixes

• Fix potential Tensor shape mismatch error in TEDPolicy and DIETClassifier.
• Update mcp version to ~1.23.0 to address security vulnerability CVE-2025-66416.
• Fix bug in CommandPayloadReader where regex matching did not account for list slots, leading to incorrect parsing of slot keys and values. Now, slot names and values are correctly extracted even if a list is provided.
• Previously, DialogueStateTracker.has_coexistence_routing_slot could incorrectly return True when the tracker was created without domain slots (i.e., using AnySlotDict), because AnySlotDict pretends all slots exist. Now, the property returns False in that case, so the routing slot is only considered present if it is actually defined in the domain.
• Fix LLM prompt template loading to raise an error instead of just printing a warning when a custom prompt file is missing or cannot be read.
• Fix AttributeError: 'str' object has no attribute 'pop' when using KnowledgeAnswerCommand with tracing enabled.
• Fix UnboundLocalError in E2ETestRunner._handle_fail_diff that caused e2e tests to crash when processing slot_was_not_set assertion failures.
• Fixed button payloads with nested parentheses (e.g., /SetSlots(slot=value)) not being parsed correctly in rasa shell. The payload was incorrectly stripped of its /SetSlots prefix, causing it to be treated as text instead of a slot-setting command.
• Add missing deprecation warning for legacy RASA_PRO_LICENSE environment variable. Update license validation error messages to reference the actual environment variable used.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.15.2] - 2025-12-04

Rasa Pro 3.15.2 (2025-12-04)

Bugfixes

• Fix agentic prompt template to access slots directly.
• Disable LLM health check for Enterprise Search Policy when generative search is disabled (i.e. use_generative_llm is False).
• Fix deduplication of collect steps in cases where the same slot was called in different flows.
• Cancel active flow before triggering internal pattern error during a custom action failure.
• Fix bug with the missing language_data module by installing langcodes with the data extra dependency.

[3.15.1] - 2025-12-02

Rasa Pro 3.15.1 (2025-12-02)

Bugfixes

• Raise validation error when duplicate slot definitions are found across domains.
• Raise validation error when a slot with an initial value set is collected by a flow collect step which sets asks_before_filling to true without having a corresponding collect utterance or custom action.
• Update pip to fix security vulnerability.
• Fix AgentToolSchema._ensure_property_types() correctly preserves structural keywords ($ref, anyOf, oneOf, etc.)

Miscellaneous internal changes

Miscellaneous internal changes.

[3.15.0] - 2025-11-26

Rasa Pro 3.15.0 (2025-11-26)

Features

• Added Langfuse integration for tracing LLM and embedding calls. All LLM-based components (Command Generators, Rephraser, Enterprise Search Policy, ReAct Sub Agent) and embedding operations are automatically traced when Langfuse is configured. Each trace includes, among other things, input/output, latency, token usage, cost, session ID, and component name.

  To get started, configure Langfuse by adding a langfuse entry to the tracing section in endpoints.yml.

  Example:

  tracing:
    - type: langfuse
      public_key: ${LANGFUSE_PUBLIC_KEY}
      private_key: ${LANGFUSE_PRIVATE_KEY}
      host: https://cloud.langfuse.com

  Custom components can override get_llm_tracing_metadata() to customize metadata.

• Added support for triggering clarification when multiple StartFlow commands are generated with no active flow. To enable this feature, set the CLARIFY_ON_MULTIPLE_START_FLOWS environment variable to True.

• Adds DTMF (Dual-Tone Multi-Frequency) input support for collect steps in flows. Voice channels can now configure DTMF collection with options for digit length, finish key, and audio input control. The feature gracefully handles non-voice channels by skipping DTMF configuration when call_state is not initialized.

• Added new CLI option -f, --e2e-failed-tests to export failed e2e tests to a file that can be directly used to re-run only those tests. It accepts an optional filename or directory path, with automatic timestamping to prevent overwriting previous test runs.

• Added support for including current datetime information by default in LLM prompts. This feature enables LLM-based components to include date and time context in their prompts, helping the model understand temporal references and provide time-aware responses.

  Components Updated:

  • CompactLLMCommandGenerator
  • SearchReadyLLMCommandGenerator
  • EnterpriseSearchPolicy
  • MCPOpenAgent (with timezone support)
  • MCPTaskAgent (with timezone support)

  Prompt Template Updates: When include_date_time is enabled, prompts include a "Date & Time Context" section showing:

  • Current date (formatted as "DD Month, YYYY")
  • Current time (formatted as "HH:MM:SS" with timezone)
  • Current day of the week

  E2E Testing Support: For deterministic testing, you can mock the datetime using the mocked_datetime slot in e2e test fixtures. The mocked datetime value must be provided in ISO 8601 format (e.g., "2024-01-15T10:30:00+00:00"). The e2e test runner validates the format and raises a ValidationError if the value is invalid.

Improvements

• Updated pattern_completed to check wether the user wants to continue the conversation or not by adding a collect step to the pattern.

• The OutputChannel is now provided to policies through the graph inputs. This enables any policy to access the output channel and send messages directly to the user, which is particularly useful for sending intermediate or filler messages.

• Allow LLM responses to be streamed to the output channel for any generative responses. This currently only applies to Enterprise Search Policy and Rephraser. Output channels that support streaming should handle duplicate message prevention when a response has already been streamed. Voice output channels (Browser Audio, Genesys, Audiocodes Stream, and Jambonz Stream) have been updated to use the streaming tokens to prepare the Text-to-Speech audio stream instead of waiting for the complete response to be generated. These changes are backward compatible; any existing channel does NOT need any additional changes. However, it can add new methods to make use of the streaming responses whenever they are available.

• Enhanced the clarification pattern with a configurable limit on the number of flows in the clarification options. Added a new slot max_clarification_options to the default_flows_for_patterns.yml with an initial value of 3 which can be changed by overriding the initial_value of the slot max_clarification_options to customize the maximum number of flows displayed during clarification.

• Enhanced the clarification pattern to handle empty clarification options. Added utter_clarification_no_options_rasa response to default_flows_for_patterns.yml, triggered when pattern_clarification receives a ClarifyCommand with no options.

• Allow e2e test results CLI flag -o, --e2e-results to be specified with a custom path. If the flag is used without a path, the default path tests/ is used.

• Enable generative responses dispatched by custom actions to be evaluated by generative_response_is_grounded and generative_response_is_relevant assertions in E2E testing. In order for the E2E testing framework to evaluate generative responses dispatched by custom actions, you must add the custom action dispatching the generative bot response to the utter_source key of the appopriate assertion in the E2E test case. For example, if you have a custom action action_generate_summary that generates a summary using a generative model, you can add the following assertion to your E2E test case:

  test_cases:
    - test_case: Generate summary for user query
      steps:
        - user: |
            Can you provide a summary of the latest news on climate change?
          assertions:
            - generative_response_is_grounded:
                threshold: 0.8
                utter_source: action_generate_summary
                ground_truth: <insert_ground_truth_summary_here>
  
            - generative_response_is_relevant:
                threshold: 0.9
                utter_source: action_generate_summary

• Allow flow_started and pattern_clarification_contains E2E testing assertions to define the operator i.e. all, any, for matching multiple flow_id values. Previously, flow_started only supported a single flow_id value, while pattern_clarification_contains defaulted to all. Introduce new format to specify the operator explicitly:

  test_cases:
    - test_case: user is asked to clarify contact-related message
      steps:
        - user: contacts
          assertions:
            - pattern_clarification_contains:
                operator: "any"
                flow_ids:
                  - remove_contact
                  - list_contacts
                  - add_contact
                  - update_contact
    - test_case: user removes a missing contact
      steps:
        - user: i want to remove a contact
          assertions:
            - flow_started:
                operator: "any"
                flow_ids:
                  - remove_contact
                  - update_contact

  The operator field can take values all or any, determining whether all specified flow_ids must match or if any one of them is sufficient.

  The previous format for these two assertions remains unchanged for backward compatibility, however it has been deprecated and will be removed in a future major release.

• Updated the default model and voice of Cartesia TTS, using sonic-3 and voice ID American English Katie.

• Move commit messages to top of the SSE

• Allow trigerring empty Clarify Command without flow options.

Bugfixes

• Run action_session_start if tracker ends with SessionEnded event.
• Fixed PostgreSQL UniqueViolation error when running an assistant with multiple Sanic workers.
• Fix Kafka producer creation failing when SASL mechanism is specified in lowercase. The SASL mechanism is now case-insensitive in the Kafka producer configuration.
• Fix issue where the validation of the assistant files continued even when the provided domain was invalid and was being loaded as empty. The training or validation command didn't exit because the final merged domain contained only the default implementations for patterns, slots and responses and therefore passed the check for being non-empty.
• Trigger pattern_internal_error in a CALM assistant when a custom action fails during execution.
• Create AWS Bedrock / Sagemaker client only if the LLM healthcheck environment variable is set. If the environment variable is not set, validate that required credentials are present.
• Update langchain-core version to ~0.3.80 to address security vulnerability CVE-2025-65106.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.14.29] - 2026-07-15

Rasa Pro 3.14.29 (2026-07-15)

Miscellaneous internal changes

Miscellaneous internal changes.

[3.14.28] - 2026-07-07

Rasa Pro 3.14.28 (2026-07-07)

Deprecations and Removals

• Breaking change: Removed the deprecated LanguageModelFeaturizer NLU component and its internal HuggingFace integration (rasa.nlu.utils.hugging_face). This removes Rasa's direct dependency on the transformers library, which was affected by CVE-2026-1839 (arbitrary code execution via a malicious checkpoint file).

  Assistants whose NLU pipeline references LanguageModelFeaturizer must migrate to a supported featurizer (for example ConveRTFeaturizer, or a CALM pipeline using LLMCommandGenerator) before upgrading. transformers is no longer a first-party component dependency. It remains a transitive dependency of gliner and is now constrained to >=5.0.0 (which contains the fix) wherever it is installed — i.e. on the gliner-based extras (full, pii). The standalone transformers extra and the sentencepiece dependency have also been removed.

[3.14.27] - 2026-06-23

Rasa Pro 3.14.27 (2026-06-23)

Bugfixes

• Bumped langchain-classic (~1.0.7, fixes CVE-2026-45134) and starlette (~1.3.1, fixes CVE-2026-54283) on 3.14.x to resolve high-severity dependency security vulnerabilities. Also raised the langchain-core floor to ~1.3.3, which is required by langchain-classic 1.0.7.

[3.14.26] - 2026-06-22

Rasa Pro 3.14.26 (2026-06-22)

Improvements

• A2A agent validation now requires TICKET_LOCK_LIFETIME to be at least as long as max_polling_time, preventing the conversation lock from expiring while A2A polling is still in progress.
• Added configure_litellm_for_rasa() in a lightweight _litellm_config module to centralize LiteLLM setup: it suppresses print-based provider-list debug footers via litellm.suppress_debug_info and sets the LiteLLM logger to WARNING, reducing log noise during LLM, embedding, and router client initialization without pulling in AWS provider dependencies.

Bugfixes

• Fixed MCP task agents exposing set_slot_* tools and listing slots in the task prompt for controlled-only or NLU-only slots in exit_if. Added a slot_mappings_allow_llm_fill helper so tool exposure, prompt rendering, and should_slot_be_set use the same rule for whether the LLM may fill a slot from its mappings.
• Fixed rasa data validate incorrectly rejecting call steps that target subagents or MCP tools when no agent configuration is loaded, instead of emitting a warning.
• Fixed suppress_logs so that concurrent async calls no longer corrupt the process-wide log level, preventing logs from disappearing permanently until pod restart.
• Refresh the AWS ElastiCache IAM presigned URL before each Redis (re)connect when within 60 seconds of expiry, so lock store and timer store operations no longer fail after >30 minutes of pro-server idleness.
• Message processing no longer crashes with ValueError: Could not find a user flow frame to cancel when an action fails (for example action_session_start returning HTTP 413) on a session whose residual dialogue stack contains only pattern frames; the internal-error pattern is pushed without attempting to cancel a non-existent user flow.
• Upgrade langsmith dependency to 0.8.5 and starlette to 1.0.1 to address CVE-2026-45134 and CVE-2026-48710.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.14.25] - 2026-06-04

Rasa Pro 3.14.25 (2026-06-04)

Bugfixes

• Fixed a spurious "Fatal error on SSL transport" traceback printed to stderr on Windows after rasa test e2e completes by explicitly closing litellm's shared httpx sessions before the event loop shuts down.

• Fixed a crash in rasa-tools where training an assistant with a sub-agent MCP server configured caused a RuntimeError: Attempted to exit cancel scope in a different task than it was entered in, crashing the stdio server and making the next tool call fail with MCP error -32000: Connection closed.

• Fix two bugs causing random crashes and resource leaks when running Rasa with --workers 2+: a pre-fork event loop left open during JWT setup, and litellm async HTTP clients not being closed on worker shutdown.

  Root cause — Sanic uses fork-based multi-worker mode (legacy=True). create_app() was calling asyncio.new_event_loop() + set_event_loop() in the main process to satisfy sanic-jwt's Initialize(), then leaving the loop open. Forked workers inherited its kqueue/epoll file descriptors, conflicting with each worker's own loop (~50% crash at startup). Separately, close_resources() never called litellm.close_litellm_async_clients(), leaving cached async HTTP clients open when workers exit.

• An explicitly configured api_key in the rasa provider model config now takes precedence over the Rasa license, allowing custom endpoints to authenticate with their own credentials.

• Fixed graph schema validation on Python 3.13+ when components use PEP 604 union type hints (e.g. Domain | None), which previously raised a false GraphSchemaValidationException.

• Transient tracker store connection errors (e.g. Postgres restart) no longer flood logs with full stack traces on each retry attempt; the error message is preserved on one line.

• Fix resume_flow() marking the wrong AgentStackFrame as resumed when a flow contains multiple sub-agent frames, which prevented run_agent() from taking its resume branch.

• Fixed multiple Langfuse session IDs being created for a single Dialogue Understanding Test (DUT) run by using the original test-case sender ID as the session ID for all per-step LLM traces.

• Fixed MCP flow tool tracing metrics when _execute_mcp_tool_call mutates the shared initial_events list in place by capturing the pre-mutation event count for output slicing, and by detecting failures from InternalErrorPatternFlowStackFrame on the dialogue stack rather than from unchanged event list length.

• Updated idna, pip and urllib3 to address security vulnerabilities.

• Update Cartesia to use the model sonic-latest from sonic-english which has been sunsetted. Also updated the API version to 2026-03-01 from 2024-06-10

• Upgrading from 3.15.x to 3.16.x no longer causes every subsequent message to fail with InvalidFlowStepIdException when a tracker was mid-flow at upgrade time and a flow step ID was renamed or removed. The flow executor now detects stale step IDs, logs a structured warning, removes the stale frame, and continues processing — rather than propagating the exception as a GraphComponentException.

[3.14.24] - 2026-06-01

Rasa Pro 3.14.24 (2026-06-01)

Bugfixes

• Fixed DIETClassifier training failures on GPU with TensorFlow 2.19 and Keras 3, where training could fail with an InvalidArgumentError about unsupported SparseReshape operations on XLA GPU JIT devices. RasaModel now inherits from tensorflow.keras.Model instead of standalone Keras 3's Model, so it uses the same Keras stack as the rest of Rasa's TensorFlow code and respects TF_USE_LEGACY_KERAS=1.

[3.14.23] - 2026-05-22

Rasa Pro 3.14.23 (2026-05-22)

Bugfixes

• Fix an interrupted sub-agent not being properly resumed after the user answers "Yes" to pattern_continue_interrupted. This is fixed by introducing a transient RESUMING agent state that sits between INTERRUPTED and WAITING_FOR_INPUT: resume_flow() now sets the agent frame to RESUMING instead of directly to WAITING_FOR_INPUT; run_agent() detects this and dispatches to _handle_resume_interrupted_agent, which replays the agent's last message and transitions the frame back to WAITING_FOR_INPUT.
• Fixed a KeyError in ContinueInterruptedPatternFlowStackFrame.from_dict that occurred when loading tracker events stored before the multi-interrupt fields were introduced. This caused repeated tracker store retries that blocked the server event loop, leading to liveness probe failures and pod restarts.

[3.14.22] - 2026-05-15

Rasa Pro 3.14.22 (2026-05-15)

Bugfixes

• Updated security-flagged dependencies:

  • litellm: ~1.81.15 → 1.84.0
  • langchain: ~0.3.27 → ~1.2.16
  • langchain-community: ~0.3.29 → ~0.4.1
  • langchain-core: ~0.3.81 → ~1.3.2
  • langsmith: ~0.6.3 → ~0.7.31
  • qdrant-client: ~1.9.1 → ~1.17.1
  • langchain-qdrant: transitive → ~1.1.0 (new direct pin)
  • langchain-openai: transitive → ~1.2.1 (new direct pin)
  • langchain-text-splitters: transitive → ~1.1.2 (new direct pin)

• Fixed a bug where sub-agent flows would end immediately after interruption resumption instead of re-invoking the agent.

  When a flow containing a sub-agent call was interrupted (e.g., by a user digression) and then resumed via pattern_continue_interrupted, the agent frame's state remained INTERRUPTED. This caused the flow executor to skip the agent call and advance to END, triggering pattern_completed prematurely.

  The agent frame state is now correctly reset to WAITING_FOR_INPUT during flow resumption, ensuring the sub-agent is re-invoked with full conversation history as expected.

• Fixed MCP tracing attribute extraction for router-based model groups so tracing no longer crashes with KeyError('provider') when instrumenting ReAct sub-agent send_message calls.

[3.14.21] - 2026-04-24

Rasa Pro 3.14.21 (2026-04-24)

Bugfixes

• Fixed a bug where the validator was mutating the allowed values list of categorical slots in-place, causing None entries to accumulate in slots used across multiple flows via subflows. This corrupted the flow retrieval FAISS embeddings during training.
• Fixed multi-agent flows looping back to an agent call when the top stack frame was interrupted instead of waiting for input, which could leave multiple agents active and fail graph execution. The loop-back shortcut now applies only when the agent frame is waiting for user input.
• Fixed validation incorrectly rejecting custom ContextualResponseRephraser subclasses configured as nlg.type in endpoints.yml. The validator now resolves the class and checks inheritance instead of only accepting the literal string "rephrase".
• Updated PyJWT, werkzeug, aiohttp, orjson, pyasn1, requests, and ujsonto resolve security vulnerabilities.
• Fixed tracing span attributes for LLM components that use model groups. Spans now carry the correct model group ID and, for router groups, the representative model's attributes (preferring OpenAI when present for prompt token counting). The llm_is_router_group attribute is also set on spans when a router group is active.

[3.14.20] - 2026-04-02

Rasa Pro 3.14.20 (2026-04-02)

Improvements

• Declared safetensors and keras as optional NLU extras (pip install 'rasa-pro[nlu]' or full), so minimal installs no longer pull them in directly. regex is also declared optional and listed under those extras for consistency with WhitespaceTokenizer, but it remains installed on most minimal installs because the base dependency tiktoken (used by the LLM stack) depends on regex.

  The affected code paths lazy-import safetensors and regex; if either package is not installed, they raise MissingDependencyException with install instructions. keras is only pulled in for the TensorFlow model stack (rasa.utils.tensorflow.models), so it is optional for the same minimal-install story as safetensors.

• Improved sub-agent call steps that use exit_if: when the same ReAct agent runs again in one conversation, slots named in exit_if are cleared so a new run does not reuse values from the last finished run. Ongoing multi-turn agent sessions (waiting for the next user message) are unchanged.

Bugfixes

• When an agent or MCP tool call returns a fatal error (e.g. internal server error), the user no longer receives two bot messages. Previously both the internal error message ("Sorry, I am having trouble with that...") and the flow-cancelled message ("Okay, stopping...") were shown. Now only the internal error message is shown; the cancel pattern is not pushed for system-initiated failures, so the flow is still marked ended for tracking but the confusing cancel utterance is omitted.

• Add rephrase endpoint validation by treating defaults-only misconfiguration as warning and user-defined rephrase misconfiguration as error.

• Upgrade langsmith dependency to 0.6.3 to address CVE-2026-25528.

• Fixed three bugs in ConcurrentRedisLockStore that caused an IndexError: deque index out of range crash in the PII anonymization and deletion cron jobs under multi-replica deployments.

  • get_lock() now returns None when no ticket keys exist in Redis (all tickets had expired), instead of returning an empty ConcurrentTicketLock object that caused downstream callers to crash.
  • save_lock() now skips the Redis write when the ticket has already expired (TTL ≤ 0), preventing a ResponseError from Redis.
  • save_lock() previously passed the absolute epoch expiry timestamp as the Redis EX TTL, resulting in ticket keys that effectively never expired (~55 years). The TTL is now correctly computed as a relative duration in seconds.

• Fixed correction reset when a user answers the active collect slot and corrects another slot in the same turn.

  CorrectSlotsCommand.create_correction_frame no longer chooses the reset step only from slots in the correction payload. It also considers the current collect-information step and picks the earlier collect step in flow order, so the stack resets to the question currently being asked instead of jumping ahead to a later slot’s collect step (which skipped validation for the new value on the active slot).

[3.14.19] - 2026-03-16

Rasa Pro 3.14.19 (2026-03-16)

Bugfixes

• Restarted task-oriented ReAct agents no longer exit immediately. When an agent is restarted, the previous run's user/assistant messages are now marked in the conversation so the model does not set slots from them; the system prompt instructs the agent to treat the current interaction as a fresh start for slot collection.
• When an A2A task has status input_required, Rasa now populates AgentOutput.structured_results from both task.artifacts and DataParts in task.status.message.parts. Previously only the text message was returned and structured data was omitted, so clients can now process artifact data when the agent is waiting for user input.
• Fix PII redaction of user or bot messages when slot values do not match the original message text. This can occur when the original messages are multi-line or TTS-style bot read-backs.
• Update setuptools to 80.10.2 to address vulnerability https://osv.dev/vulnerability/DEBIAN-CVE-2026-23949. Replaced randomname with duoname (no dependencies) library.

[3.14.18] - 2026-03-11

Rasa Pro 3.14.18 (2026-03-11)

Deprecations and Removals

• Removed an explicit deprecation warning for the license varible RASA_PRO_LICENSE to provide clarity that we will continue to support both this and the newer RASA_LICENSE variable for the forseeable future.

Bugfixes

• When a flow had Agent A → flow steps → Agent B and the user restarted Agent A while Agent B was already started (then interrupted), execution after the restarted Agent A completed would jump back to Agent B and skip the flow steps between A and B. Flow steps between agents are now executed correctly, and the previously interrupted Agent B is resumed instead of started from scratch.

• Fixes MCP tool result handling to now process results that contain only structuredContent (and no or empty content). Previously, an empty content field caused the result to be skipped and slots were not set from structuredContent.

• Downgraded the A2A polling "waiting to poll again" log from ERROR to INFO so normal polling no longer triggers false alerts.

• SQL tracker store: update() now supports a content-only path when the timestamp-based delete removes no rows (e.g. anonymization: same timestamps, content changed). In that case, the store either updates existing event rows in place where content differs or performs a full replace if event counts differ, so anonymized content is persisted correctly in a single atomic transaction.

  Anonymization cron job: the privacy manager now uses update(updated_tracker) for anonymization instead of delete-then-save. SQL tracker store behavior is aligned with MongoDB, Redis and DynamoDB (overwrite semantics), and anonymization with SQL completes in one atomic operation without a separate delete/save sequence.

• Prevent JsonPatchConflict exceptions raised when tracker is split in sub-sessions during PII anonymization cron jobs. Replace usage of a utility function that was recreating tracker objects using sub-sessions with an approach retrieving list of events for each sub-session instead.

[3.14.17] - 2026-03-03

Rasa Pro 3.14.17 (2026-03-03)

Bugfixes

• • Kafka event broker background polling thread caused test hangs and didn't allow the process to exit: The background poll thread is now a daemon thread, so the process can terminate when the broker is not closed explicitly (e.g. in tests or after an abrupt exit). The close() method uses a 5-second join timeout so it does not block forever if the poll thread is stuck (e.g. when the broker is unreachable).
  • Background polling: The poll loop now runs for the whole lifetime of the broker, even before the producer is created. When the producer is not ready, the loop sleeps briefly instead of exiting. This allows the IAM OAuth callback to be triggered when using SASL_SSL with AWS IAM (MSK), so tokens can be refreshed as the library calls poll().
  • Connection keepalive (optional): New broker options socket_keepalive_enable, reconnect_backoff_ms, reconnect_backoff_max_ms, and topic_metadata_refresh_interval_ms let you reduce idle connection drops. When socket_keepalive_enable is true, these are passed to the underlying producer.

• Fixes race conditions in PII cron jobs by requiring a LockStore for BackgroundPrivacyManager and acquiring a per-sender_id lock while anonymization/deletion jobs read-modify-write trackers.

  Implemented measures to defend against JSON patch issues and ensure correct event order.

  • Deletion cron job: When reconstructing the tracker from retained events after deleting old sessions, the deletion job now catches JsonPatchException and JsonPointerException (e.g. from invalid or inconsistent dialogue stack updates). On failure, the tracker is left unchanged and an error is logged instead of overwriting or deleting data, so no data loss occurs. Users can resolve stack-related issues by appending a Restarted event to the tracker if needed.
  • Anonymization cron job: Events from already-anonymized, processed, and ineligible sessions are now sorted by timestamp before rebuilding the tracker, with original index used as a tie-breaker when timestamps are identical. This keeps replay order chronologically consistent and ensures dialogue stack updates and other order-dependent events are applied in the correct sequence.

• ReAct agents now filter conversation events to user and bot utterances before building assistant and user role messages, and the default context limit is 10 utterance messages. Previously, the last 20 events were taken first and then filtered, so fewer user and assistant messages were included in the context.

[3.14.16] - 2026-02-26

Rasa Pro 3.14.16 (2026-02-26)

Bugfixes

• Upgraded litellm to 1.81.15 so that strict: True can be passed to Bedrock and tool calling has the same guarantees as with OpenAI. Upgraded openai to >=2.8.0 to satisfy litellm's dependency.
• Unquoted environment variable references for MCP and A2A OAuth credentials (client_id, client_secret) in endpoints.yml are now accepted. Previously, training failed unless these values were wrapped in double quotes (e.g. "${MCP_CLIENT_SECRET}").
• Move MCP task agent exit condition evaluation to after process_output, so that SlotSet events added by custom process_output implementations are considered when checking exit_if conditions.
• The MCP task agent now includes slot changes in the agent output when the state is INPUT_REQUIRED or when max iterations is reached. Previously, slots set via set_slot tools were updated in memory but not forwarded in the output, so they were not persisted until exit conditions were met.

[3.14.15] - 2026-02-23

Rasa Pro 3.14.15 (2026-02-23)

Bugfixes

• Remove config file content from endpoint read success logs to prevent sensitive data exposure.

• Update protobuf to v5.29.6 to address CVE-2026-0994. Update wheel to v0.46.3 to address CVE-2026-24049.

• E2E fixture resolution and conftest hierarchy:

  Fixture resolution now uses a conftest-style hierarchy, with local overrides taking precedence over global fixtures. This is a change from the previous behavior where all fixtures were merged into a single list, which could lead to silent data loss if duplicate fixture names were used. Now, every test case has its own resolved set of fixtures, and duplicate fixture names are allowed when the intent is "override".

  Details:

  • Conftest-style hierarchy: Fixtures can be defined at root or folder level (e.g. conftest.yml / conftest.yaml); all e2e tests under that path see them.
  • Single-file run parity: Running one test file resolves fixtures the same way as when that file is run as part of the full suite (global/conftest fixtures are loaded for that file’s path).
  • Runtime namespace: Each test case has its own resolved set of fixtures; there is no shared mutable fixture state between tests. Duplicate fixture names in different files are allowed when the intent is “override” (no need to remove or rename for full-suite runs).
  • Override semantics: Local (file-level) fixtures override folder-level; folder-level overrides root. Within a single file, duplicate fixture names remain an error.

[3.14.14] - 2026-02-12

Rasa Pro 3.14.14 (2026-02-12)

• Upgrade litellm to 1.80.0.
• Fix Enterprise Search Policy triggering utter_ask_rephrase instead of utter_no_relevant_answer_found when the vector store search returns no matching documents. The pattern_cannot_handle now correctly receives the cannot_handle_no_relevant_answer as a context.reason in all cases where no relevant answer is found, not only when the optional relevancy check rejects the answer.

[3.14.13] - 2026-02-05

Rasa Pro 3.14.13 (2026-02-05)

Bugfixes

• Upgrade Keras to 3.12.1 to address CVE-2026-0897.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.14.12] - 2026-01-26

Rasa Pro 3.14.12 (2026-01-26)

Bugfixes

• Update azure-core to 1.38.0 to address CVE-2026-21226.

[3.14.11] - 2026-01-21

Rasa Pro 3.14.11 (2026-01-21)

Bugfixes

• Fixed e2e test coverage report to correctly track coverage per flow. Each flow now appears as a separate entry with accurate coverage percentages. Additionally, call/link steps and collect steps with prefilled slots are now properly marked as visited.
• Throw error when duplicate fixtures are found in the same file or across files.
• Fix OAuth2AuthStrategy to use URL-encoded parameters with Basic Authentication to fetch access token

[3.14.10] - 2026-01-15

Rasa Pro 3.14.10 (2026-01-15)

Bugfixes

• Fixed an issue where storing documents in FAISS individually could hit the token limit. Now, documents are stored in batches to avoid exceeding the token limit.
• SessionEnded does not reset the tracker anymore.
• Updated werkzeug, aiohttp, filelock, fonttools, marshmallow, urllib3 and langchain-coreto resolve security vulnerabilities.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.14.9] - 2025-12-29

Rasa Pro 3.14.9 (2025-12-29)

Bugfixes

• Include response button titles in conversation history in prompts of LLM based components.
• Fix OpenTelemetry exporter Invalid type <class 'NoneType'> of value None error when recording the request duration metric for requests made from Rasa server to the custom action server. This error occurred because the url parameter was not being passed to the method that records the request duration metric, resulting in a NoneType value.
• Fix DUT fails with AttributeError when running with custom CommandGenerator

[3.14.8] - 2025-12-19

Rasa Pro 3.14.8 (2025-12-19)

Bugfixes

• Fixed token expiration validation failing on servers running in non-UTC timezones. Token expiration checks now use timezone-aware UTC datetimes consistently, preventing premature "access token expired" errors on systems in timezones like UTC+8.

[3.14.7] - 2025-12-11

Rasa Pro 3.14.7 (2025-12-11)

Bugfixes

• Fix potential Tensor shape mismatch error in TEDPolicy and DIETClassifier.
• Update mcp version to ~1.23.0 to address security vulnerability CVE-2025-66416.
• Fix bug in CommandPayloadReader where regex matching did not account for list slots, leading to incorrect parsing of slot keys and values. Now, slot names and values are correctly extracted even if a list is provided.
• Previously, DialogueStateTracker.has_coexistence_routing_slot could incorrectly return True when the tracker was created without domain slots (i.e., using AnySlotDict), because AnySlotDict pretends all slots exist. Now, the property returns False in that case, so the routing slot is only considered present if it is actually defined in the domain.
• Fix LLM prompt template loading to raise an error instead of just printing a warning when a custom prompt file is missing or cannot be read.
• Fix AttributeError: 'str' object has no attribute 'pop' when using KnowledgeAnswerCommand with tracing enabled.
• Fix UnboundLocalError in E2ETestRunner._handle_fail_diff that caused e2e tests to crash when processing slot_was_not_set assertion failures.
• Fixed button payloads with nested parentheses (e.g., /SetSlots(slot=value)) not being parsed correctly in rasa shell. The payload was incorrectly stripped of its /SetSlots prefix, causing it to be treated as text instead of a slot-setting command.
• Add missing deprecation warning for legacy RASA_PRO_LICENSE environment variable. Update license validation error messages to reference the actual environment variable used.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.14.6] - 2025-12-05

Rasa Pro 3.14.6 (2025-12-05)

Bugfixes

• Fix agentic prompt template to access slots directly.
• Disable LLM health check for Enterprise Search Policy when generative search is disabled (i.e. use_generative_llm is False).
• Fix deduplication of collect steps in cases where the same slot was called in different flows.
• Cancel active flow before triggering internal pattern error during a custom action failure.
• Fix bug with the missing language_data module by installing langcodes with the data extra dependency.

[3.14.5] - 2025-12-02

Rasa Pro 3.14.5 (2025-12-02)

Bugfixes

• Update pip to fix security vulnerability.
• Fix AgentToolSchema._ensure_property_types() correctly preserves structural keywords ($ref, anyOf, oneOf, etc.)

[3.14.4] - 2025-11-27

Rasa Pro 3.14.4 (2025-11-27)

Bugfixes

• Fixed patten-continue-interrupted running out of order before linked flows.
• Fix rasa studio upload timeouts by enabling TCP keep-alive with platform-specific socket options to maintain stable connections.
• Remove action_metadata tracing span attribute from EnterpriseSearchPolicy instrumentation to prevent PII leakages. Add new environment variable RASA_TRACING_DEBUGGING_ENABLED to enable adding action_metadata to EnterpriseSearchPolicy spans for debugging purposes. By default, this variable is set to false to ensure PII is not logged in production environments.
• Fixed PostgreSQL UniqueViolation error when running an assistant with multiple Sanic workers.
• Fix Kafka producer creation failing when SASL mechanism is specified in lowercase. The SASL mechanism is now case-insensitive in the Kafka producer configuration.
• Fix issue where the validation of the assistant files continued even when the provided domain was invalid and was being loaded as empty. The training or validation command didn't exit because the final merged domain contained only the default implementations for patterns, slots and responses and therefore passed the check for being non-empty.
• Trigger pattern_internal_error in a CALM assistant when a custom action fails during execution.
• Create AWS Bedrock / Sagemaker client only if the LLM healthcheck environment variable is set. If the environment variable is not set, validate that required credentials are present.
• Update langchain-core version to ~0.3.80 to address security vulnerability CVE-2025-65106.
• Raise validation error when duplicate slot definitions are found across domains.
• Raise validation error when a slot with an initial value set is collected by a flow collect step which sets asks_before_filling to true without having a corresponding collect utterance or custom action.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.14.3] - 2025-11-13

Rasa Pro 3.14.3 (2025-11-13)

Miscellaneous internal changes

Miscellaneous internal changes.

[3.14.2] - 2025-10-30

Rasa Pro 3.14.2 (2025-10-30)

Improvements

• Add new environment variable LOG_LEVEL_PYMONGO to control the logging level of PyMongo dependency of Rasa. This can be useful to reduce the verbosity of logs. Default value is INFO.

  The logging level of PyMongo can also be set via the LOG_LEVEL_LIBRARIES environment variable, which provides the default logging level for a selection of third-party libraries used by Rasa. If both variables are set, LOG_LEVEL_PYMONGO takes precedence.

Bugfixes

• Clean up duplicated ChitChatAnswerCommands in command processor.

  Ensure that one CannotHandleCommand remains if only multiple CannotHandleCommands were present.

• LLM request timeouts are now enforced at the event loop level using asyncio.wait_for.

  Previously, timeout values configured in endpoints.yml could be overridden by the HTTP client's internal timeout behavior. This fix ensures that when a specific timeout value is configured for LLM requests, the request respects that exact timing regardless of the underlying HTTP client implementation.

• Fixed an issue where duplicate collect steps in a flow could cause rendering problems, such as exceeding token limits. This occurred when a flow called another flow multiple times. Now, each collect step is listed only once when retrieving all collect steps for a flow.

• If a FloatSlot does not have min and max values set in the domain, the slot will not be validated against any range. If the slot defines an initial value, as well as min and max values, the initial value will be validated against the range and an error will be raised if the initial value is out of range. Enhance run-time validation for new values assigned to FloatSlot instances, ensuring they fall within the defined min and max range if these are set. If min and max are not set, no range validation is performed.

• Fixed bug preventing deployment parameter from being used in generative response LLM judge configuration.

• Fixed bug where persisted_slots defined in called flows were incorrectly reset when the parent flow ended, unless they were also explicitly defined in the parent flow. Persisted slots now only need to be defined in the called flow to remain persisted after the parent flow ends.

• Fixes the prompt template resolution logic in the CompactLLMCommandGenerator and SearchReadyLLMCommandGenerator classes.

• When action_clean_stack is used in a user flow, allow pattern_completed to be triggered even if there are pattern flows at the bottom of the dialogue stack below the top user flow frame. Previously, pattern_completed would only trigger if there were no other frames below the top user flow frame. This assumes that the action_clean_stack is the last action called in the user flow.

• Update pip version used in Dockerfile to 23.* to address security vulnerability CVE-2023-5752. Update python-socketio version to 5.14.0 to address security vulnerability CVE-2025-61765.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.14.1] - 2025-10-10

Rasa Pro 3.14.1 (2025-10-10)

Miscellaneous internal changes

Miscellaneous internal changes.

[3.14.0] - 2025-10-09

Rasa Pro 3.14.0 (2025-10-09)

Features

• Added the possibility to call an agent from within a flow via a call step. We support two types of ReAct style agents:

  1. An open ended exploratory agent - Here the user’s goal is loosely defined, for example researching about stocks to invest in. There is no pre-defined criteria which can be used to know when the goal is reached. Hence the agent itself needs to figure out when it has helped the user to the best of its capabilities.

     flows:
       stock_investment_research:
         description: helps research and analyze stock investment options
         steps:
           - call: stock_explorer  # runs until agent signals completion

  2. A task specific agent - Here we can come up with conditions, which when true, signals the completion of a user goal. For example, helping the user find an appointment slot based on their preferences - you know the agent can exit once all user’s preferences have been met and they have agreed to a slot.

     flows:
       book_appointment:
         description: helps users book appointments
         steps:
           - call: ticket_agent
             exit_if:
               - slots.chosen_appointment is not null
               - slots.booking_completed is True**
           - collect: user_confirmation

  To configure the agents you need to create a folder structure like this:

  your_project/
  ├── config.yml
  ├── domain/
  ├── data/flows/
  └── sub_agents/
      └── stock_explorer/
          ├── config.yml  # mandatory
          └── prompt_template.jinja2  # optional

  The config.yml file for each agent is mandatory and should look, for example, like this:

  # Basic agent information
  agent:
    name: stock_explorer # name of the agent
    protocol: RASA # protocol for its connections either RASA or A2A
    description: "Agent that helps users research and analyze stock options" # a brief description used by the command generator to continue / stop the agent and the agent during its execution
  
  # Core configuration
  configuration:
    llm:  # (optional) Same format as other Rasa LLM configs
      type: openai
      model: gpt-4
    prompt_template: sub_agents/stock_explorer/prompt_template.jinja2  # (optional) prompt template file containing a customized prompt
    timeout: 30  # (optional) seconds before timing out
    max_retries: 3  # (optional) connection retry attempts
  
  # MCP server connections
  connections:
    mcp_servers:
      - name: trade_server
        include_tools:  # optional: specify which tools to include
          - find_symbol
          - get_company_news
          - apply_technical_analysis
          - fetch_live_price
        exclude_tools:  # optional: tools to exclude; helpful to not allow the agent to execute critical tools.
          - place_buy_order
          - view_positions

• Added the possibility to call an MCP tool directly from a flow step via a call step.

  Example:

    - call: place_buy_order  # MCP tool name
      mcp_server: trade_server # MCP server where tool is available
      mapping:
        input:
        - param: ticker_symbol    # tool parameter name
          slot: stock_name     # slot to send as value
        - param: quantity
          slot: order_quantity
        output:
        - slot: order_status    # slot to store results
          result_key: result.structuredContent.order_status.success

  The MCP server needs to be defined in the endpoints.yml file

  mcp_servers:
    - name: trade_server
      url: http://localhost:8080
      type: http

• We updated the inspector

  • to highlight when a call step is calling an agent or an MCP tool using a little icon in the flow view.
  • and added an additional agent widget that shows the available agents and their corresponding status in the conversation.

• Whenever agents are configured, the SearchReadyLLMCommandGenerator and the CompactLLMCommandGenerator will use new default prompt templates that contain new agent specific commands (RestartAgent and ContinueAgent) and instructions.

• Adds Interruption Handling Support (beta) in Voice Stream Channels, namely Browser Audio, Twilio Media Streams and Jambonz Stream. Interruptions are identified from partial transcripts sent by the Automatic Speech Recogintion (ASR) Engine. Interruption Handling is disabled by defaulat and can be configured using the optional interruptions key in credentials.yml.

  browser_audio:
    server_url: localhost
    interruptions:
      enabled: True
      min_words: 3
    asr:
      name: "deepgram"
    tts:
      name: cartesia

  The configuration property min_words (default 3) requires the partial transcript to have at least 3 words to interrupt the bot. This is a rudimentary way to filter backchannels (brief responses like "hmm", "yeah", "ok").

  Interruptions can be disabled for certain responses using the domain response property allow_interruptions (default true). Example,

  responses:
    utter_current_balance:
      - text: You still have {current_balance} in your account.
        allow_interruptions: false

• Added the following new events to capture the state of any agent used within a conversation:

  • AgentStarted
  • AgentCompleted
  • AgentCancelled
  • AgentResumed
  • AgentInterrupted

• Added support for Redis Cluster mode in the tracker store for horizontal scaling and cloud compatibility.

  Added support for Redis Sentinel mode in the tracker store for high availability with automatic failover.

  Added 3 new configuration parameters to RedisTrackerStore:

  • deployment_mode: String value specifying Redis deployment type ("standard", "cluster", or "sentinel"). Defaults to "standard".
  • endpoints: List of strings in "host:port" format defining cluster nodes or sentinel instances.
  • sentinel_service: String value specifying the sentinel service name to connect to. Only used in sentinel mode, defaults to "mymaster".

• Add support for IAM authentication to AWS RDS SQL tracker store. To enable this feature:

  • set the IAM_CLOUD_PROVIDER environment variable to aws.
  • set the AWS_DEFAULT_REGION environment variable to your desired AWS region.
  • if you want to enforce SSL verification, you can set the SQL_TRACKER_STORE_SSL_MODE environment variable (for example, to verify-full or verify-ca) and provide the path to the CA certificate using the SQL_TRACKER_STORE_SSL_ROOT_CERTIFICATE environment variable.

• Add support for IAM authentication to AWS Managed Streaming for Kafka event broker. This allows secure authentication to an AWS MSK cluster without the need for static credentials. To enable this feature:

  • set the IAM_CLOUD_PROVIDER environment variable to aws.
  • set the AWS_DEFAULT_REGION environment variable to your desired AWS region.
  • ensure that your application has the necessary IAM permissions to access the AWS Managed Service Kafka cluster.
  • ensure the topic is created on the MSK cluster.
  • use the SASL_SSL security protocol and OAUTHBEARER SASL mechanism in your Kafka configuration in endpoints.yml.
  • download the root certificate from Amazon Trust Services: https://www.amazontrust.com/repository/ and provide its path to the ssl_cafile event broker yaml property.

• Added support for Redis Cluster mode in RedisLockStore for horizontal scaling and cloud compatibility.

  Added support for Redis Sentinel mode in RedisLockStore for high availability with automatic failover.

  Added 3 new configuration parameters to RedisLockStore:

  • deployment_mode: String value specifying Redis deployment type ("standard", "cluster", or "sentinel"). Defaults to "standard".
  • endpoints: List of strings in "host:port" format defining cluster nodes or sentinel instances.
  • sentinel_service: String value specifying the sentinel service name to connect to. Only used in sentinel mode, defaults to "mymaster".

• Added support for Redis Cluster mode in ConcurrentRedisLockStore for horizontal scaling and cloud compatibility.

  Added support for Redis Sentinel mode in ConcurrentRedisLockStore for high availability with automatic failover.

  Added 3 new configuration parameters to ConcurrentRedisLockStore:

  • deployment_mode: String value specifying Redis deployment type ("standard", "cluster", or "sentinel"). Defaults to "standard".
  • endpoints: List of strings in "host:port" format defining cluster nodes or sentinel instances.
  • sentinel_service: String value specifying the sentinel service name to connect to. Only used in sentinel mode, defaults to "mymaster".

• Add support for IAM authentication to AWS ElastiCache for Redis lock store. This allows secure authentication to an AWS ElastiCache without the need for static credentials. To enable this feature:

  • set the IAM_CLOUD_PROVIDER environment variable to aws.
  • set the AWS_DEFAULT_REGION environment variable to your desired AWS region.
  • download the root certificate from Amazon Trust Services and provide its path to the ssl_ca_certs lock store yaml property.
  • if enabling cluster mode, set the AWS_ELASTICACHE_CLUSTER_NAME environment variable to your ElastiCache cluster name.
  • ensure that your application has the necessary IAM permissions to access the AWS ElastiCache for Redis cluster.

• Default silence timeout is now configurable per channel. Default silence timeout is read from channel configuration in credentials file. Example:

  audiocodes_stream:
      server_url: "humble-arguably-stork.ngrok-free.app"
      asr:
          name: deepgram
      tts:
          name: cartesia
  
      silence_timeout: 5

  Silence timeout at collect step is also configurable per channel:

  flows:
      - name: example_flow
        steps:
          - collect:
              silence_timeout:
                  - audiocodes_stream: 5

  If not configured, default silence timeout is 7 seconds for each channel.

• Adds support for Deepgram TTS Adds browser_audio channel to the credentials.yml in tutorial template

Improvements

• In case a StartFlowCommand for a flow, that is already on the stack, is predicted, we now resume the flow and interrupt the current active flow instead of ignoring the StartFlowCommand.

• Updated pattern_continue_interrupted to ask the user if they want to proceed with any of the interrupted flows before resuming that particular flow.

• We have isolated the dependencies that are only required for NLU components. These include:

  • transformers (version: "~4.38.2")
  • tensorflow (version: "^2.19.0, only available for "python_version < '3.12')
  • tensorflow-text (version "^2.19.0", only available for "python_version < '3.12')
  • tensorflow-hub (version: "^0.13.0", only available for "python_version < '3.12')
  • tensorflow-io-gcs-filesystem (version: "==0.31" for sys_platform == 'win32', version: "==0.34" for sys_platform == 'linux', version: "==0.34" for sys_platform == 'linux' for "sys_platform == 'darwin' and platform_machine != 'arm64'; all only available for "python_version < '3.12')
  • tensorflow-metal (version: "^1.2.0", only available for "python_version < '3.12')
  • tf-keras (version: "^2.15.0", only available for "python_version < '3.12')
  • spacy (version: "^3.5.4")
  • sentencepiece (version: "~0.1.99", only available for "python_version < '3.12')
  • skops (version: "~0.13.0")
  • mitie (version: "^0.7.36", added for consistency)
  • jieba (version: ">=0.42.1, <0.43")
  • sklearn-crfsuite (version: "~0.5.0")

  The dependencies for those components are now in their own extra, called nlu and can be installed via: poetry install --extras nlu, pip install 'rasa-pro[nlu]' or uv pip install 'rasa-pro[nlu]'.

• We added dependencies for agent orchestration, mcp ("~1.12.0") and a2a-sdk ("~0.3.4"), to the list of default dependencies. As a result, they get installed via: poetry install, pip install 'rasa-pro' or uv pip install 'rasa-pro'.

  To ensure compatibility between a2a-sdk and tensorflow, tensorflow and its related dependencies were updated. These include:

  • tensorflow (from "2.14.1" to "^2.19.0")
  • tensorflow-text (from "2.14.0" to ^2.19.0)
  • tensorflow-metal (from "1.1.0" to "^1.2.0")
  • instead of keras ("2.14.0"), we now use tf-keras ("^2.15.0")

  Additionally, tensorflow dependencies that are incompatible with the upgraded tensorflow version were removed. These include: tensorflow-intel, tensorflow-cpu-aws, tensorflow-macos.

  Because tensorflow and related packages were upgraded, tensorflow code was also updated. As a consequence, the incremental training feature is not supported anymore. This is because tf-keras ("^2.15.0") does not allow a compiled model to be updated.

• We updated the supported Python versions:

  • Dropped support for Python 3.9.
  • Added support for Python 3.12 and Python 3.13. So, whilst we previously supported python = ">=3.9.2,<3.12", we now support python = ">=3.10.0,<3.14".

  To ensure compatibility with the newly supported Python versions, existing dependencies had to be updated. These include:

  • protobuf (from "~4.25.8" to "~5.29.5")
  • importlib-resources (from "6.1.3" to "^6.5.2")
  • opentelemetry-sdk (from "~1.16.0" to "~1.33.0")
  • opentelemetry-exporter-jaeger ("~1.16.0") - has been removed
  • opentelemetry-exporter-otlp (from "~1.16.0" to "~1.33.0")
  • opentelemetry-api (from "~1.16.0" to "~1.33.0")
  • pep440-version-utils - only available for "python_version < '3.13'"
  • pymilvus (from ">=2.4.1,<2.4.2" to "^2.6.1")
  • numpy (from "~1.26.4" to "~2.1.3")
  • scipy ("~1.13.1" for "python_version < '3.12'"; "~1.14.0" for "python_version >= '3.12'")
  • tensorflow (version: "^2.19.0, only available for "python_version < '3.12')
  • tensorflow-text (version "^2.19.0", only available for "python_version < '3.12')
  • tensorflow-hub (version: "^0.13.0", only available for "python_version < '3.12')
  • tensorflow-io-gcs-filesystem (version: "==0.31" for sys_platform == 'win32', version: "==0.34" for sys_platform == 'linux', version: "==0.34" for sys_platform == 'linux' for "sys_platform == 'darwin' and platform_machine != 'arm64'; all only available for "python_version < '3.12')
  • tensorflow-metal (version: "^1.2.0", only available for "python_version < '3.12')
  • tf-keras (version: "^2.15.0", only available for "python_version < '3.12')
  • sentencepiece (version: "~0.1.99", only available for "python_version < '3.12')

  Note that tensorflow and related dependencies are only supported for Python < 3.12. As a result, components requiring those dependencies are not available for Python >= 3.12. These are: DIETClassifier , TEDPolicy , UnexpecTEDIntentPolicy , ResponseSelector , ConveRTFeaturizer and LanguageModelFeaturizer .

• We have isolated the dependencies required for channel connectors. These include:

  • fbmessenger (version: "~6.0.0")
  • twilio (version: "~9.7.2")
  • webexteamssdk (version: ">=1.6.1,<1.7.0")
  • mattermostwrapper (version: "~2.2")
  • rocketchat_API (version: ">=1.32.0,<1.33.0")
  • aiogram (version: "~3.22.0")
  • slack-sdk (version: "~3.36.0")
  • cvg-python-sdk (version: "^0.5.1")

  The dependencies for those channels are now in their own extra, called channels and can be installed via: poetry install --extras channels, pip install 'rasa-pro[channels]' or uv pip install 'rasa-pro[channels]'. Note that the dependencies for the channels browser_audio, studio_chat, socketIO and rest are not included and remain in the main list of dependencies.

• Validation errors now raise meaningful exceptions instead of calling sys.exit(1), producing clearer and more actionable log messages while allowing custom error handling.

• Engine-related modules now raise structured exceptions instead of calling sys.exit(1) or print_error_and_exit, providing clearer, more actionable log messages and enabling custom error handling.

• Enable connection to AWS services for LLMs (e.g. Bedrock, Sagemaker) via multiple additional methods to environment variables, for example by using IAM roles or AWS credentials file.

• Relocate latency display in Inspector

• Implement Redis connection factory class to handle different Redis deployment modes. This class abstracts Redis connection creation and automatically selects the appropriate redis-py client (Redis, RedisCluster, or Sentinel) based on configuration.

  • In standard mode, connects to a single Redis instance.
  • In cluster mode, connects to a Redis Cluster using the provided endpoints.
  • In sentinel mode, connects to a Redis Sentinel setup for high availability.

• Metadata of action_listen event in the tracker contains the execution times for Command Processor and Prediction Loop for the previous turn. This information is present in the key execution_times in the metadata and the times are in milliseconds.

• Rasa Voice Inspector (browser_audio channel) used to require a Rasa License with a specific voice scope. This scope requirement has been dropped.

• Add new environment variables for each AWS service integration (RDS, ElastiCache for Redis, MSK) that indicates whether to use IAM authentication when connecting to the service:

  • KAFKA_MSK_AWS_IAM_ENABLED - set to true to enable IAM authentication for MSK connections.
  • RDS_SQL_DB_AWS_IAM_ENABLED - set to true to enable IAM authentication for RDS connections.
  • ELASTICACHE_REDIS_AWS_IAM_ENABLED - set to true to enable IAM authentication for ElastiCache for Redis connections.

Bugfixes

• Pass all flows to find_updated_flows to avoid creating a HandleCodeChangeCommand in situations where flows were not updated.
• The .devcontainer docker-compose file now uses the new docker.io/bitnamilegacy repository for images, rather than the old docker.io/bitnami. This change was made in response to Bitnami's announcement that they will be putting all of their public Helm Chart container images behind a paywall starting August 28th, 2025. Existing public images are moved to the new legacy repository - bitnamilegacy - which is only intended for short-term migration purposes.
• In case an agents fails with an fatal error or returns an unkown state, we now cancel the current active flow next to triggering pattern_internal_error.
• Bugfix for Jambonz Stream channel Websocket URL path which resulted in failed Websocket Connections
• Fixed the contextual response rephraser to use and update the correct translated response text for the current language, instead of falling back to the default response.
• Refactored validate_argument_paths to accept list values and aggregate all missing paths before exiting.
• Fix expansion of referenced environment variables in endpoints.yml during Bedrock model config validation.
• Enabled the data argument to support both string and list inputs, normalizing to a list for consistent handling.
• Fixed model loading failures on Windows systems running recent Python versions (3.9.23+, 3.10.18+, 3.11.13+) due to tarfile security fix incompatibility with Windows long path prefix.
• Fixes the silence handling bug in Audiocodes Stream Channel where consecutive bot responses could trip silence timeout pattern while the bot is speaking. Audiocodes Stream channel now forecefully cancels the silence timeout watcher whenever it sends the bot response audio.
• Use correct formatting method for messages in completion and acompletion functions of LiteLLMRouterLLMClient.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.13.24] - 2026-02-23

Rasa Pro 3.13.24 (2026-02-23)

Bugfixes

• E2E fixture resolution and conftest hierarchy:

  Fixture resolution now uses a conftest-style hierarchy, with local overrides taking precedence over global fixtures. This is a change from the previous behavior where all fixtures were merged into a single list, which could lead to silent data loss if duplicate fixture names were used. Now, every test case has its own resolved set of fixtures, and duplicate fixture names are allowed when the intent is "override".

  Details:

  • Conftest-style hierarchy: Fixtures can be defined at root or folder level (e.g. conftest.yml / conftest.yaml); all e2e tests under that path see them.
  • Single-file run parity: Running one test file resolves fixtures the same way as when that file is run as part of the full suite (global/conftest fixtures are loaded for that file’s path).
  • Runtime namespace: Each test case has its own resolved set of fixtures; there is no shared mutable fixture state between tests. Duplicate fixture names in different files are allowed when the intent is “override” (no need to remove or rename for full-suite runs).
  • Override semantics: Local (file-level) fixtures override folder-level; folder-level overrides root. Within a single file, duplicate fixture names remain an error.

[3.13.23] - 2026-02-12

Rasa Pro 3.13.23 (2026-02-12)

Improvements

• Flows can now link to pattern_search (e.g. to trigger RAG or branch on knowledge-based search).

Bugfixes

• Fixed e2e test coverage report to correctly track coverage per flow. Each flow now appears as a separate entry with accurate coverage percentages. Additionally, call/link steps and collect steps with prefilled slots are now properly marked as visited.
• Upgrade litellm to 1.80.0.
• Fix Enterprise Search Policy triggering utter_ask_rephrase instead of utter_no_relevant_answer_found when the vector store search returns no matching documents. The pattern_cannot_handle now correctly receives the cannot_handle_no_relevant_answer as a context.reason in all cases where no relevant answer is found, not only when the optional relevancy check rejects the answer.

[3.13.22] - 2026-01-15

Rasa Pro 3.13.22 (2026-01-15)

Bugfixes

• Fixed an issue where storing documents in FAISS individually could hit the token limit. Now, documents are stored in batches to avoid exceeding the token limit.
• Throw error when duplicate fixtures are found in the same file or across files.
• SessionEnded does not reset the tracker anymore.
• Updated werkzeug, filelock, fonttools, marshmallow, urllib3 and langchain-coreto resolve security vulnerabilities.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.13.21] - 2025-12-29

Rasa Pro 3.13.21 (2025-12-29)

Bugfixes

• Include response button titles in conversation history in prompts of LLM based components.
• Fix OpenTelemetry exporter Invalid type <class 'NoneType'> of value None error when recording the request duration metric for requests made from Rasa server to the custom action server. This error occurred because the url parameter was not being passed to the method that records the request duration metric, resulting in a NoneType value.
• Fix DUT fails with AttributeError when running with custom CommandGenerator

[3.13.20] - 2025-12-19

Rasa Pro 3.13.20 (2025-12-19)

Bugfixes

• Fixed token expiration validation failing on servers running in non-UTC timezones. Token expiration checks now use timezone-aware UTC datetimes consistently, preventing premature "access token expired" errors on systems in timezones like UTC+8.

[3.13.19] - 2025-12-11

Rasa Pro 3.13.19 (2025-12-11)

Bugfixes

• Fix bug in CommandPayloadReader where regex matching did not account for list slots, leading to incorrect parsing of slot keys and values. Now, slot names and values are correctly extracted even if a list is provided.
• Previously, DialogueStateTracker.has_coexistence_routing_slot could incorrectly return True when the tracker was created without domain slots (i.e., using AnySlotDict), because AnySlotDict pretends all slots exist. Now, the property returns False in that case, so the routing slot is only considered present if it is actually defined in the domain.
• Fix LLM prompt template loading to raise an error instead of just printing a warning when a custom prompt file is missing or cannot be read.
• Fix AttributeError: 'str' object has no attribute 'pop' when using KnowledgeAnswerCommand with tracing enabled.
• Fix UnboundLocalError in E2ETestRunner._handle_fail_diff that caused e2e tests to crash when processing slot_was_not_set assertion failures.
• Fixed button payloads with nested parentheses (e.g., /SetSlots(slot=value)) not being parsed correctly in rasa shell. The payload was incorrectly stripped of its /SetSlots prefix, causing it to be treated as text instead of a slot-setting command.
• Add missing deprecation warning for legacy RASA_PRO_LICENSE environment variable. Update license validation error messages to reference the actual environment variable used.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.13.18] - 2025-12-04

Rasa Pro 3.13.18 (2025-12-04)

Bugfixes

• Update pip to fix security vulnerability.
• Disable LLM health check for Enterprise Search Policy when generative search is disabled (i.e. use_generative_llm is False).
• Fix deduplication of collect steps in cases where the same slot was called in different flows.
• Cancel active flow before triggering internal pattern error during a custom action failure.
• Fix bug with the missing language_data module by installing langcodes with the data extra dependency.

[3.13.17] - 2025-11-27

Rasa Pro 3.13.17 (2025-11-27)

Bugfixes

• Fixed PostgreSQL UniqueViolation error when running an assistant with multiple Sanic workers.
• Fix Kafka producer creation failing when SASL mechanism is specified in lowercase. The SASL mechanism is now case-insensitive in the Kafka producer configuration.
• Fix issue where the validation of the assistant files continued even when the provided domain was invalid and was being loaded as empty. The training or validation command didn't exit because the final merged domain contained only the default implementations for patterns, slots and responses and therefore passed the check for being non-empty.
• Trigger pattern_internal_error in a CALM assistant when a custom action fails during execution.
• Create AWS Bedrock / Sagemaker client only if the LLM healthcheck environment variable is set. If the environment variable is not set, validate that required credentials are present.
• Update langchain-core version to ~0.3.80 to address security vulnerability CVE-2025-65106.
• Raise validation error when duplicate slot definitions are found across domains.
• Raise validation error when a slot with an initial value set is collected by a flow collect step which sets asks_before_filling to true without having a corresponding collect utterance or custom action.

[3.13.16] - 2025-11-21

Rasa Pro 3.13.16 (2025-11-21)

Bugfixes

• Fixed patten-continue-interrupted running out of order before linked flows.
• Fix rasa studio upload timeouts by enabling TCP keep-alive with platform-specific socket options to maintain stable connections.
• Remove action_metadata tracing span attribute from EnterpriseSearchPolicy instrumentation to prevent PII leakages. Add new environment variable RASA_TRACING_DEBUGGING_ENABLED to enable adding action_metadata to EnterpriseSearchPolicy spans for debugging purposes. By default, this variable is set to false to ensure PII is not logged in production environments.

[3.13.15] - 2025-11-13

Rasa Pro 3.13.15 (2025-11-13)

Miscellaneous internal changes

Miscellaneous internal changes.

[3.13.14] - 2025-10-30

Rasa Pro 3.13.14 (2025-10-30)

Improvements

• Add new environment variable LOG_LEVEL_PYMONGO to control the logging level of PyMongo dependency of Rasa. This can be useful to reduce the verbosity of logs. Default value is INFO.

  The logging level of PyMongo can also be set via the LOG_LEVEL_LIBRARIES environment variable, which provides the default logging level for a selection of third-party libraries used by Rasa. If both variables are set, LOG_LEVEL_PYMONGO takes precedence.

Bugfixes

• Clean up duplicated ChitChatAnswerCommands in command processor.

  Ensure that one CannotHandleCommand remains if only multiple CannotHandleCommands were present.

• LLM request timeouts are now enforced at the event loop level using asyncio.wait_for.

  Previously, timeout values configured in endpoints.yml could be overridden by the HTTP client's internal timeout behavior. This fix ensures that when a specific timeout value is configured for LLM requests, the request respects that exact timing regardless of the underlying HTTP client implementation.

• Fixed an issue where duplicate collect steps in a flow could cause rendering problems, such as exceeding token limits. This occurred when a flow called another flow multiple times. Now, each collect step is listed only once when retrieving all collect steps for a flow.

• If a FloatSlot does not have min and max values set in the domain, the slot will not be validated against any range. If the slot defines an initial value, as well as min and max values, the initial value will be validated against the range and an error will be raised if the initial value is out of range. Enhance run-time validation for new values assigned to FloatSlot instances, ensuring they fall within the defined min and max range if these are set. If min and max are not set, no range validation is performed.

• Fixed bug preventing deployment parameter from being used in generative response LLM judge configuration.

• Fixed bug where persisted_slots defined in called flows were incorrectly reset when the parent flow ended, unless they were also explicitly defined in the parent flow. Persisted slots now only need to be defined in the called flow to remain persisted after the parent flow ends.

• Fixes the prompt template resolution logic in the CompactLLMCommandGenerator and SearchReadyLLMCommandGenerator classes.

• When action_clean_stack is used in a user flow, allow pattern_completed to be triggered even if there are pattern flows at the bottom of the dialogue stack below the top user flow frame. Previously, pattern_completed would only trigger if there were no other frames below the top user flow frame. This assumes that the action_clean_stack is the last action called in the user flow.

• Update pip version used in Dockerfile to 23.* to address security vulnerability CVE-2023-5752. Update python-socketio version to 5.14.0 to address security vulnerability CVE-2025-61765.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.13.13] - 2025-10-13

Rasa Pro 3.13.13 (2025-10-13)

Bugfixes

• Use correct formatting method for messages in completion and acompletion functions of LiteLLMRouterLLMClient.

[3.13.12] - 2025-09-17

Rasa Pro 3.13.12 (2025-09-17)

Improvements

• Accept either RASA_PRO_LICENSE or RASA_LICENSE as a valid Environment Variable for providing the Rasa License. Also deprecates RASA_PRO_LICENSE Environment Variable.

Bugfixes

• Fixed conversation hanging when using direct action execution with invalid actions module by deferring module validation until action execution time, allowing proper exception handling in the processor.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.13.11] - 2025-09-15

Rasa Pro 3.13.11 (2025-09-15)

No significant changes.

[3.13.10] - 2025-09-12

Rasa Pro 3.13.10 (2025-09-12)

Bugfixes

• Fixes the silence handling bug in Audiocodes Stream Channel where consecutive bot responses could trip silence timeout pattern while the bot is speaking. Audiocodes Stream channel now forcefully cancels the silence timeout watcher whenever it sends the bot response audio.
• Update langchain to 0.3.27 and langchain-community to 0.3.29 to fix CVE-2025-6984 vulnerability in langchain-community version 0.2.19.

[3.13.9] - 2025-09-03

Rasa Pro 3.13.9 (2025-09-03)

Bugfixes

• Upgrade skops to version 0.13.0 and requests to version 2.32.5 to fix vulnerabilities.
• Fixed flow retrieval to skip vector store population when there are no flows to embed.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.13.8] - 2025-08-26

Rasa Pro 3.13.8 (2025-08-26)

Improvements

• Validation errors now raise meaningful exceptions instead of calling sys.exit(1), producing clearer and more actionable log messages while allowing custom error handling.
• Enable connection to AWS services for LLMs (e.g. Bedrock, Sagemaker) via multiple additional methods to environment variables, for example by using IAM roles or AWS credentials file.

Bugfixes

• The .devcontainer docker-compose file now uses the new docker.io/bitnamilegacy repository for images, rather than the old docker.io/bitnami. This change was made in response to Bitnami's announcement that they will be putting all of their public Helm Chart container images behind a paywall starting August 28th, 2025. Existing public images are moved to the new legacy repository - bitnamilegacy - which is only intended for short-term migration purposes.
• Fix expansion of referenced environment variables in endpoints.yml during Bedrock model config validation.
• Fixed model loading failures on Windows systems running recent Python versions (3.9.23+, 3.10.18+, 3.11.13+) due to tarfile security fix incompatibility with Windows long path prefix.

[3.13.7] - 2025-08-15

Rasa Pro 3.13.7 (2025-08-15)

Bugfixes

• Don't tigger a slot correction for slots that are currently set to None as None counts as empty value.
• Enabled the data argument to support both string and list inputs, normalizing to a list for consistent handling.

[3.13.6] - 2025-08-08

Rasa Pro 3.13.6 (2025-08-08)

Bugfixes

• Fixed the contextual response rephraser to use and update the correct translated response text for the current language, instead of falling back to the default response.
• Refactored validate_argument_paths to accept list values and aggregate all missing paths before exiting.

[3.13.5] - 2025-07-31

Rasa Pro 3.13.5 (2025-07-31)

Bugfixes

• Fix correction of slots:

  • Slots can only be corrected in case they belong to any flow on the stack and the slot to be corrected is part of a collect step in any of those flows.
  • A correction of a slot should be applied if the flow that is about to start is using this slot.

• Upgrade axios to fix security vulnerability.

• Fix issue where called flows could not set slots of their parent flow.

[3.13.4] - 2025-07-23

Rasa Pro 3.13.4 (2025-07-23)

Improvements

• Engine-related modules now raise structured exceptions instead of calling sys.exit(1) or print_error_and_exit, providing clearer, more actionable log messages and enabling custom error handling.

Bugfixes

• Fix validation of the FAISS documents folder to ensure it correctly discovers files in a recursive directory structure.
• Updated Inspector dependent packages (vite, and @adobe/css-tools) to address security vulnerabilities.
• Fixed bug preventing model_group from being used with embeddings in generative response LLM judge configuration.

[3.13.3] - 2025-07-17

Rasa Pro 3.13.3 (2025-07-17)

Bugfixes

• Allowed NLG servers to return None for the text property and ensured custom response data is properly retained.

[3.13.2] - 2025-07-16

Rasa Pro 3.13.2 (2025-07-16)

Bugfixes

• Pass all flows to find_updated_flows to avoid creating a HandleCodeChangeCommand in situations where flows were not updated.
• Bugfix for Jambonz Stream channel Websocket URL path which resulted in failed Websocket Connections

[3.13.1] - 2025-07-14

Rasa Pro 3.13.1 (2025-07-14)

Bugfixes

• Fix issues with bot not giving feedback for slot corrections.
• Fixed bot speaking state management in Audiocodes Stream channel. This made the assistant unable to handle user silences Added periodic keepAlive messages to deepgram, this interval can be configured with keep_alive_interval parameter

[3.13.0] - 2025-07-07

Rasa Pro 3.13.0 (2025-07-07)

Deprecations and Removals

• Deprecate IntentlessPolicy and schedule for removal in Rasa 4.0.0.
• Removed monitor_silence parameter from Voice Channel configuration. Silence Monitoring is now enabled by default. It can be configured by changing the value of Global Silence Timeout
• Remove pre-CALM PII management capability originally released in Rasa Plus 3.6.0. Remove presidio, faker and pycountry dependencies from Rasa Pro.

Features

• Introducing the SearchReadyLLMCommandGenerator component, an enhancement over the CompactLLMCommandGenerator. This new component improves the triggering accuracy of KnowledgeAnswerCommand and should be used when the EnterpriseSearchPolicy is added to the pipeline. By default, this new component does not trigger the ChitChatAnswerCommand and HumanHandoffCommand. Handling small talk and chit-chat conversations is now delegated to the EnterpriseSearchPolicy.

  To incorporate the SearchReadyLLMCommandGenerator into your pipeline, simply add the following:

  pipeline:

  ...
  - name: SearchReadyLLMCommandGenerator
  ...

• Implement Privacy Filter capability to detect PII in 3 supported events (SlotSet, BotUttered, and UserUttered) and anonymise PII from the event data.

  The PII detection takes a tiered approach:

  • the first tier represents a slot-based approach: the sensitive data is stored in a slot whose name is defined in the privacy YAML configuration.
  • the second optional tier uses a default GliNer model to detect PII that is not captured by the slot-based approach.

  The anonymisation of PII is done by redacting or replacing the sensitive data with a placeholder. These anonymisation rules are also defined in the privacy YAML configuration.

• EnterpriseSearchPolicy can now assess the relevancy of the generated answer. By default, the policy does not check the relevancy of the generated answer. But you enable the relevancy check by setting check_relevancy to true in the policy configuration.

    policies:
      - name: FlowPolicy
      - name: EnterpriseSearchPolicy
        ...
        check_relevancy: true # by default this is false

  If the relevancy check is enabled, the policy will evaluate the generated answer and determine if it is relevant to the user's query. In case the answer is relevant, the generated answer will be returned to the user. If the answer is not relevant, the policy will fallback to pattern_cannot_handle to handle the situation.

  pattern_cannot_handle:
    description: |
      Conversation repair flow for addressing failed command generation scenarios
    name: pattern cannot handle
    steps:
      - noop: true
        next:
            ... # other cases
          # Fallback to handle cases where the generated answer is not relevant
              - if: "'{{context.reason}}' = 'cannot_handle_no_relevant_answer'"
              then:
                - action: utter_no_relevant_answer_found
                  next: "END"

• Implement privacy manager class which manages privacy-related tasks in the background.

  This class handles the anonymization and deletion of sensitive information in dialogue state trackers stored in the tracker store, as well as the streaming of anonymized events to event brokers. It uses background schedulers to periodically run these tasks and processes trackers from a queue to ensure that sensitive information is handled in a timely manner.

• Add support for jambonz_stream voice-stream channel. Log level of websockets library is set to ERROR by default, it can be changed by the environment variable LOG_LEVEL_LIBRARIES.

• Remove the beta feature flag check from the RepeatBotMessagesCommand. The RASA_PRO_BETA_REPEAT_COMMAND environment variable is no longer needed as the feature is GA in 3.13.0.

Improvements

• Coverage report feature can now be used independently of RASA_PRO_BETA_FINE_TUNING_RECIPE feature flag.

• Update default Embedding model from text-embedding-ada-002 to text-embedding-3-large. Update LLMBasedRouter, IntentlessPolicy and ContextualResponseRephraser default models to use gpt-4o-2024-11-20 instead of gpt-3.5-turbo. Update the EnterpriseSearchPolicy to use gpt-4.1-mini-2025-04-14 instead of gpt-3.5-turbo.

• Update MultistepCommandGenerator to use gpt-4o, specifically gpt-4o-2024-11-20, instead of gpt-3.5-turbo as gpt-3.5-turbo will be deprecated on July 16, 2025. Add deprecation warning for SingleStepCommandGenerator; leave current default model as gpt-4-0613. Adapt tests for CommandGenerators to use gpt-4o instead of gpt-3.5-turbo due to upcoming model deprecation. Update LLMJudgeModel and Fine tuning Conversation Rephraser to use gpt-4.1-mini, specifically gpt-4.1-mini-2025-04-14, instead of gpt-4o-mini, as gpt-4o-mini will be deprecated on September 15, 2025.

• Make CALM template the default for rasa init.

• Redis lock store now accepts host property from endpoints config. Property url is marked as deprecated for Redis lock store.

• Added support for basic authentication in Twilio channels (Voice Ready and Voice Streaming). This allows users to authenticate their Twilio channels using basic authentication credentials, enhancing security and access control for voice communication. To use this feature, set username and password in the Twilio channel configuration.

  credentials.yaml

  twilio_voice:
      username: your_username
      password: your_password
      ...
  
  twilio_media_streams:
      username: your_username
      password: your_password
      ...

  At Twilio, configure the webhook URL to include the basic authentication credentials:

  # twilio voice webhook
  https://<username>:<password>@yourdomain.com/webhooks/twilio_voice/webhook
  
  # twilio media streams webhook
  https://<username>:<password>@yourdomain.com/webhooks/twilio_media_streams/webhook

• Added two endpoints to the model service for retrieving the assistant’s default configuration and the default project template used by the assistant.

• All conversations on Twilio Media Streams, Audiocodes Stream, Genesys channel ends with the message /session_end. This applies to both the conversation ended by user and the assistant.

• Refactor Voice Inspector to use AudioWorklet Web API

• Added new CLI commands to support project-level linking and granular push/pull workflows for Rasa Studio assistants:

  • Introduced rasa studio link to associate a local project with a specific Studio assistant.
  • Added rasa studio pull and rasa studio push, with subcommands for granular resource updates (config, endpoints) between local and Studio assistants.

• Implement delete method for InMemoryTrackerStore,AuthRetryTrackerStore, AwaitableTrackerStore and FailSafeTrackerStore subclasses.

• Implement delete method for MongoTrackerStore.

• Implement delete method for SQLTrackerStore: this method accepts sender_id and deletes the corresponding tracker from the store if it exists.

• Implement delete method for DynamoTrackerStore: this method accepts sender_id and deletes the corresponding tracker from the store if it exists.

• Implement delete method for RedisTrackerStore.

• Update KafkaEventBroker YAML config to accept 2 new parameters:

  • stream_pii: boolean (default: true). If set to false, the broker will not publish un-anonymised events to the configured topic.
  • anonymization_topics: list of strings (default: []). If set, the broker will publish anonymised events to the configured topics when the PII management capability is enabled.

• Add 2 new PII configuration parameters to PikaEventBroker:

  • stream_pii: Boolean flag to control whether or not to publish un-anonymised events to the configured queues. If set to False, un-anonymised events won't be published to RabbitMQ. Defaults to True for backwards compatibility.
  • anonymization_queues: List of queue names that should receive anonymized events with PII removed. Defaults to an empty list ([]).

• Add a new anonymized_at timestamp field to the Rasa Pro events supported by the PII management capability:

  • user
  • slot
  • bot This field indicates when the PII data in the event was anonymized. The field is set to null if the PII data has not been anonymized.

• Add support for reading privacy configuration key from endpoints file to enable PII management capability.

• Add new DELETE /conversations/<conversation_id>/tracker endpoint that allows deletion of tracker data for a specific conversation.

• Cleaned up potential sources of PII from info, exception and error logs.

• Allow default patterns to link to pattern_chitchat.

• Cleaned up potential sources of PII from warning logs.

• Add a warning log to alert users when exporting tracker data that contains unanonymized events.

• Remove beta feature flag for pypred predicate usage in conditional response variations. Mark this functionality for general availability (GA).

• Updated the default behavior of pattern_chitchat. With the deprecation of IntentlessPolicy, pattern_chitchat now defaults to responding with utter_cannot_handle instead of triggering action_trigger_chitchat.

• PII deletion job now performs a single database transaction in the case of trackers with multiple sessions, where only some sessions are eligible for deletion. The deletion job overwrites the serialized tracker record with the retained events only. This ensures that the tracker store is updated atomically, preventing any potential tracker data loss in case of Rasa Pro server crashes or other issues during the deletion job execution.

Bugfixes

• Fixes a bug where fine-tuning data generation raises a FineTuningDataPreparationException for test cases without assertions that end with one/multiple user utterance/s, as opposed to one/multiple bot utterance/s.

  For these types of test cases, the fine-tuning data generation transcript now contains all test case utterances up to but excluding the last user utterance(s) as the test case does not specify the corresponding, expected bot response(s).

• allowed for usage of litellm model prefixes that do not end in '<provider>/'

• Fixes a bug in Inspector that raised a TypeError when serialising numpy.float64 from Tracker

• • Fixes an issue with prompt rendering where minified JSON structures were displayed without properly escaping newlines, tabs, and quotes.
  • Introduced a new Jinja filter to_json_encoded_string that escapes newlines (\n), tabs (\t), and quotes (\") for safe JSON rendering. to_json_encoded_string filter preserves other special characters (e.g., umlauts) without encoding them.
  • Updated the default prompts for gpt-4o and claude-sonnet-3.5

• Fix intermittent crashes on Inspector app when Enterprise Search or Chitchat Policies were triggered

• Add channel name to UserMessage created by the Audiocodes channel.

• Reduced redundant log entries when reading prompt templates by contextualizing them. Added source component and method metadata to logs, and changed prompt-loading logs triggered from fingerprint_addon to use DEBUG level.

• Prompts for rephrased messages and conversations are now rendered using agent, eliminating errors that occurred when string replacements broke after prompt updates.

• Fix retrieval of latest tracker session from DynamoTrackerStore.

• Files generated by the finetuning data generation pipeline are now encoded in UTF-8, allowing characters such as German umlauts (ä, ö, ü) to render correctly.

• Fix an issue where the SetSlot and Clarify command value was parsed incorrectly if a newline character immediately followed the value argument in the LLM output.

• Fixed the repeat action to include all messages of the last turn in collect steps.

• • Fix an issue where running rasa inspect would always set the route_session_to_calm slot to True, even when no user-triggered commands were present. This caused incorrect routing to CALM, bypassing the logic of the router.
  • Fix a regression in non-sticky routing where sessions intended for the NLU were incorrectly routed to CALM when the router predicted NoopCommand().

• Fix validation and improve error messages for the documents source directory when FAISS vector store is configured for the Enterprise Search Policy.

• Prevent slot correction when the slot is already set to the desired value.

• Fallback to CannotHandle command when the slot predicted by the LLM is not defined in the domain.

• Fix tracker store PII background jobs not updating the InMemoryTrackerStore reference stored by the Agent.

• Fix potential KeyError in EnterpriseSearchPolicy citation post-processing when the source does not have the correct citation. Improve the citation post-processing logic to handle additional edge cases.

• Fix issue with PrivacyManager unable to retrieve trackers from Redis tracker store during its deletion background job, because the Redis tracker store prefix was added twice during retrieval.

• Fixed anonymization failing for FloatSlots when user-provided integer values don't match the stored float representation during text replacement.

• Fix Redis tracker store saving tracker with prefix added twice during the anonymization background job.

• Fix DynamoDB tracker store overwriting tracker with the latest session in case of multiple sessions saved prior.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.12.42] - 2025-12-11

Rasa Pro 3.12.42 (2025-12-11)

Bugfixes

• Fix bug in CommandPayloadReader where regex matching did not account for list slots, leading to incorrect parsing of slot keys and values. Now, slot names and values are correctly extracted even if a list is provided.
• Previously, DialogueStateTracker.has_coexistence_routing_slot could incorrectly return True when the tracker was created without domain slots (i.e., using AnySlotDict), because AnySlotDict pretends all slots exist. Now, the property returns False in that case, so the routing slot is only considered present if it is actually defined in the domain.
• Fix LLM prompt template loading to raise an error instead of just printing a warning when a custom prompt file is missing or cannot be read.
• Fix AttributeError: 'str' object has no attribute 'pop' when using KnowledgeAnswerCommand with tracing enabled.
• Fix UnboundLocalError in E2ETestRunner._handle_fail_diff that caused e2e tests to crash when processing slot_was_not_set assertion failures.
• Fixed button payloads with nested parentheses (e.g., /SetSlots(slot=value)) not being parsed correctly in rasa shell. The payload was incorrectly stripped of its /SetSlots prefix, causing it to be treated as text instead of a slot-setting command.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.12.41] - 2025-12-04

Rasa Pro 3.12.41 (2025-12-04)

Bugfixes

• Update pip to fix security vulnerability.
• Disable LLM health check for Enterprise Search Policy when generative search is disabled (i.e. use_generative_llm is False).
• Fix deduplication of collect steps in cases where the same slot was called in different flows.
• Cancel active flow before triggering internal pattern error during a custom action failure.
• Fix bug with the missing language_data module by installing langcodes with the data extra dependency.

[3.12.40] - 2025-11-27

Rasa Pro 3.12.40 (2025-11-27)

Bugfixes

• Fixed PostgreSQL UniqueViolation error when running an assistant with multiple Sanic workers.
• Fix issue where the validation of the assistant files continued even when the provided domain was invalid and was being loaded as empty. The training or validation command didn't exit because the final merged domain contained only the default implementations for patterns, slots and responses and therefore passed the check for being non-empty.
• Trigger pattern_internal_error in a CALM assistant when a custom action fails during execution.
• Update langchain-core version to ~0.3.80 to address security vulnerability CVE-2025-65106.
• Raise validation error when duplicate slot definitions are found across domains.
• Raise validation error when a slot with an initial value set is collected by a flow collect step which sets asks_before_filling to true without having a corresponding collect utterance or custom action.

[3.12.39] - 2025-11-21

Rasa Pro 3.12.39 (2025-11-21)

Bugfixes

• Fixed patten-continue-interrupted running out of order before linked flows.
• Remove action_metadata tracing span attribute from EnterpriseSearchPolicy instrumentation to prevent PII leakages. Add new environment variable RASA_TRACING_DEBUGGING_ENABLED to enable adding action_metadata to EnterpriseSearchPolicy spans for debugging purposes. By default, this variable is set to false to ensure PII is not logged in production environments.
• Fix Kafka producer creation failing when SASL mechanism is specified in lowercase. The SASL mechanism is now case-insensitive in the Kafka producer configuration.

[3.12.38] - 2025-11-13

Rasa Pro 3.12.38 (2025-11-13)

Miscellaneous internal changes

Miscellaneous internal changes.

[3.12.37] - 2025-10-30

Rasa Pro 3.12.37 (2025-10-30)

Bugfixes

• Clean up duplicated ChitChatAnswerCommands in command processor.

  Ensure that one CannotHandleCommand remains if only multiple CannotHandleCommands were present.

• LLM request timeouts are now enforced at the event loop level using asyncio.wait_for.

  Previously, timeout values configured in endpoints.yml could be overridden by the HTTP client's internal timeout behavior. This fix ensures that when a specific timeout value is configured for LLM requests, the request respects that exact timing regardless of the underlying HTTP client implementation.

• Fixed an issue where duplicate collect steps in a flow could cause rendering problems, such as exceeding token limits. This occurred when a flow called another flow multiple times. Now, each collect step is listed only once when retrieving all collect steps for a flow.

• Fixes the prompt template resolution logic in the CompactLLMCommandGenerator.

• When action_clean_stack is used in a user flow, allow pattern_completed to be triggered even if there are pattern flows at the bottom of the dialogue stack below the top user flow frame. Previously, pattern_completed would only trigger if there were no other frames below the top user flow frame. This assumes that the action_clean_stack is the last action called in the user flow.

• Update pip version used in Dockerfile to 23.* to address security vulnerability CVE-2023-5752.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.12.36] - 2025-10-27

Rasa Pro 3.12.36 (2025-10-27)

Miscellaneous internal changes

Miscellaneous internal changes.

[3.12.35] - 2025-10-21

Rasa Pro 3.12.35 (2025-10-21)

Improvements

• Add new environment variable LOG_LEVEL_PYMONGO to control the logging level of PyMongo dependency of Rasa. This can be useful to reduce the verbosity of logs. Default value is INFO.

  The logging level of PyMongo can also be set via the LOG_LEVEL_LIBRARIES environment variable, which provides the default logging level for a selection of third-party libraries used by Rasa. If both variables are set, LOG_LEVEL_PYMONGO takes precedence.

Bugfixes

• If a FloatSlot does not have min and max values set in the domain, the slot will not be validated against any range. If the slot defines an initial value, as well as min and max values, the initial value will be validated against the range and an error will be raised if the initial value is out of range. Enhance run-time validation for new values assigned to FloatSlot instances, ensuring they fall within the defined min and max range if these are set. If min and max are not set, no range validation is performed.
• Fixed bug preventing deployment parameter from being used in generative response LLM judge configuration.
• Fixed bug where persisted_slots defined in called flows were incorrectly reset when the parent flow ended, unless they were also explicitly defined in the parent flow. Persisted slots now only need to be defined in the called flow to remain persisted after the parent flow ends.

[3.12.34] - 2025-10-13

Rasa Pro 3.12.34 (2025-10-13)

Bugfixes

• Fixed conversation hanging when using direct action execution with invalid actions module by deferring module validation until action execution time, allowing proper exception handling in the processor.
• Use correct formatting method for messages in completion and acompletion functions of LiteLLMRouterLLMClient.

[3.12.33] - 2025-09-12

Rasa Pro 3.12.33 (2025-09-12)

Bugfixes

• Fixes the silence handling bug in Audiocodes Stream Channel where consecutive bot responses could trip silence timeout pattern while the bot is speaking. Audiocodes Stream channel now forcefully cancels the silence timeout watcher whenever it sends the bot response audio.
• Fixed flow retrieval to skip vector store population when there are no flows to embed.

[3.12.32] - 2025-09-03

Rasa Pro 3.12.32 (2025-09-03)

Bugfixes

• Upgrade skops to version 0.13.0 and requests to version 2.32.5 to fix vulnerabilities.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.12.31] - 2025-08-26

Rasa Pro 3.12.31 (2025-08-26)

Bugfixes

• The .devcontainer docker-compose file now uses the new docker.io/bitnamilegacy repository for images, rather than the old docker.io/bitnami. This change was made in response to Bitnami's announcement that they will be putting all of their public Helm Chart container images behind a paywall starting August 28th, 2025. Existing public images are moved to the new legacy repository - bitnamilegacy - which is only intended for short-term migration purposes.
• Fixed model loading failures on Windows systems running recent Python versions (3.9.23+, 3.10.18+, 3.11.13+) due to tarfile security fix incompatibility with Windows long path prefix.

[3.12.30] - 2025-08-19

Rasa Pro 3.12.30 (2025-08-19)

Bugfixes

• Don't tigger a slot correction for slots that are currently set to None as None counts as empty value.

[3.12.29] - 2025-07-31

Rasa Pro 3.12.29 (2025-07-31)

Bugfixes

• Fix correction of slots:

  • Slots can only be corrected in case they belong to any flow on the stack and the slot to be corrected is part of a collect step in any of those flows.
  • A correction of a slot should be applied if the flow that is about to start is using this slot.

[3.12.28] - 2025-07-29

Rasa Pro 3.12.28 (2025-07-29)

Bugfixes

• Upgrade axios to fix security vulnerability.
• Fix issue where called flows could not set slots of their parent flow.

[3.12.27] - 2025-07-23

Rasa Pro 3.12.27 (2025-07-23)

Bugfixes

• Fix validation of the FAISS documents folder to ensure it correctly discovers files in a recursive directory structure.
• Updated Inspector dependent packages (vite, and @adobe/css-tools) to address security vulnerabilities.
• Fixed bug preventing model_group from being used with embeddings in generative response LLM judge configuration.

[3.12.26] - 2025-07-17

Rasa Pro 3.12.26 (2025-07-17)

Bugfixes

• Allowed NLG servers to return None for the text property and ensured custom response data is properly retained.

[3.12.25] - 2025-07-16

Rasa Pro 3.12.25 (2025-07-16)

Bugfixes

• Pass all flows to find_updated_flows to avoid creating a HandleCodeChangeCommand in situations where flows were not updated.

[3.12.24] - 2025-07-14

Rasa Pro 3.12.24 (2025-07-14)

Bugfixes

• Fixed the repeat action to include all messages of the last turn in collect steps.
• Fix issues with bot not giving feedback for slot corrections.
• Fixed bot speaking state management in Audiocodes Stream channel. This made the assistant unable to handle user silences Added periodic keepAlive messages to deepgram, this interval can be configured with keep_alive_interval parameter

[3.12.23] - 2025-07-08

Rasa Pro 3.12.23 (2025-07-08)

Bugfixes

• Reverted fix for custom multilingual output payloads being overwritten.

[3.12.22] - 2025-07-07

Rasa Pro 3.12.22 (2025-07-07)

No significant changes.

[3.12.21] - 2025-07-03

Rasa Pro 3.12.21 (2025-07-03)

Bugfixes

• Fixes a bug where fine-tuning data generation raises a FineTuningDataPreparationException for test cases without assertions that end with one/multiple user utterance/s, as opposed to one/multiple bot utterance/s.

  For these types of test cases, the fine-tuning data generation transcript now contains all test case utterances up to but excluding the last user utterance(s) as the test case does not specify the corresponding, expected bot response(s).

• Fix validation and improve error messages for the documents source directory when FAISS vector store is configured for the Enterprise Search Policy.

• Prevent slot correction when the slot is already set to the desired value.

• Fallback to CannotHandle command when the slot predicted by the LLM is not defined in the domain.

• Fix potential KeyError in EnterpriseSearchPolicy citation post-processing when the source does not have the correct citation. Improve the citation post-processing logic to handle additional edge cases.

[3.12.20] - 2025-06-24

Rasa Pro 3.12.20 (2025-06-24)

Bugfixes

• Flows now traverse called and linked flows, including nested and branching called / linked flows. As a result, E2E coverage reports include any linked and called flows triggered by the flow being tested.
• Fixed a bug in Genesys and Audiocodes Stream channels where conversations used a placeholder value default as the Sender ID. Added a new method VoiceInputChannel.get_sender_id(call_parameters) that returns the platform's call_id as the Sender ID. This ensures each conversation has a unique identifier based on the call ID from the respective platform. Channels can override this method to customize Sender ID generation.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.12.19] - 2025-06-18

Rasa Pro 3.12.19 (2025-06-18)

Bugfixes

• Fix issues where linked flows could not be cancelled and slots collected within linked flows could not be prefilled.
• Fix InvalidFlowStepIdException thrown when a bot is restarted with a retrained model which contains an update to the step order in a given active flow. Once retrained and restarted, the bot will now correctly handle the updated step order by triggering pattern_code_change.

[3.12.18] - 2025-06-12

Rasa Pro 3.12.18 (2025-06-12)

Bugfixes

• Ensure that old step ID formats (without the flow ID prefix) can be loaded without raising an InvalidFlowStepIdException in newer Rasa versions that expect the flow ID prefix.
• • Fix an issue where running rasa inspect would always set the route_session_to_calm slot to True, even when no user-triggered commands were present. This caused incorrect routing to CALM, bypassing the logic of the router.
  • Fix a regression in non-sticky routing where sessions intended for the NLU were incorrectly routed to CALM when the router predicted NoopCommand().
• Enable slot prefilling in patterns.

[3.12.17] - 2025-06-05

Rasa Pro 3.12.17 (2025-06-05)

Bugfixes

• Fix an issue where the SetSlot and Clarify command value was parsed incorrectly if a newline character immediately followed the value argument in the LLM output.

[3.12.16] - 2025-06-03

Rasa Pro 3.12.16 (2025-06-03)

Bugfixes

• Make domain an optional argument in CommandProcessorComponent. This change addresses a potential TypeError that could occur when loading a model trained without providing domain as a required argument. By making domain optional, models trained with older configurations or without a domain component will now load correctly without errors.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.12.15] - 2025-06-02

Rasa Pro 3.12.15 (2025-06-02)

Improvements

:

Bugfixes

• Add turn_wrapper to count multiple utterances by bot/user as single turn rather than individual turns. Always include last user utterance in rephraser prompt's conversation history.
• Remove StoryGraphProvider from the prediction graph in case IntentlessPolicy is not present to reduce the loading time of the bot in cases where the IntentlessPolicy is not used and a lot of stories are present.

[3.12.14] - 2025-05-28

Rasa Pro 3.12.14 (2025-05-28)

Improvements

• • Updates the parameter name from max_tokens, which is deprecated by OpenAI, to max_completion_tokens. The old max_tokens is not supported for the o models.
  • Exposes LiteLLM's drop_params parameter for LLM configurations.

Bugfixes

• • Fixes default config initialization for the IntentlessPolicy.
• Prompts for rephrased messages and conversations are now rendered using agent, eliminating errors that occurred when string replacements broke after prompt updates.
• Fix parsing of escape characters in translation of LLM prediction to SetSlotCommand.
• Files generated by the finetuning data generation pipeline are now encoded in UTF-8, allowing characters such as German umlauts (ä, ö, ü) to render correctly.

[3.12.13] - 2025-05-19

Rasa Pro 3.12.13 (2025-05-19)

Bugfixes

• The Clarify (syntax used by SingleStepLLMCommandGenerator) / disambiguate flows (syntax used by CompactLLMCommandGenerator) command will now parse flow names with dashes.
• Fix remote model download when models are stored in a path, not in the root of the remote storage. Add new training CLI param --remote-root-only that can be used by the model service to store the model in the root of the remote storage. Propagate this parameter to the persistor's persist method. Simplify persistor code when retrieving models by downloading the model to the target path directly rather than copying the downloaded model to the target path. This also improved testability.
• When inspector is not used, root server path should output: Hello from Rasa: <version>.. When inspector is used, root server path should output HTML page with a link to the path on which inspector can be reached.
• Change the default value of minimize_num_calls in the config of LLM-based command generators to True.

[3.12.12] - 2025-05-15

Rasa Pro 3.12.12 (2025-05-15)

Improvements

• Improved CRFEntityExtractor persistence and loading methods to improve model loading times.

Bugfixes

• Bumps aiohttp to 3.10.x, sentry-sdk to 2.8.x. Also bumps the locked versions for urllib3 and h11

[3.12.11] - 2025-05-14

Rasa Pro 3.12.11 (2025-05-14)

No significant changes.

[3.12.10] - 2025-05-08

Rasa Pro 3.12.10 (2025-05-08)

Bugfixes

• Fix issues in Audiocodes Channel Connector. The values in user_phone and bot_phone available in session_started_metadata are swapped to correctly map to calller and callee respectively. Fixed evening handling where events without the key "parameters" raised an exception.
• Filtered out only None values from the response payload to avoid removing valid empty values.

[3.12.9] - 2025-05-06

Rasa Pro 3.12.9 (2025-05-06)

Bugfixes

• Implemented backtracking and corrected the recursion logic in the all-paths generation for a flow. Removed the need to deepcopy the step_ids_visited set on each branch within _handle_links, which prevents the coverage report from freezing due to hitting the recursion limit.
• Upgrade openai and litellm dependencies to fix found vulnerabilities in litellm.

[3.12.8] - 2025-04-30

Rasa Pro 3.12.8 (2025-04-30)

Bugfixes

• Reduced redundant log entries when reading prompt templates by contextualizing them. Added source component and method metadata to logs, and changed prompt-loading logs triggered from fingerprint_addon to use DEBUG level.
• Add support for auth_token to Rasa Inspector.
• Fixed issue where a slot mapping referencing a form in the active_loop slot mapping conditions that wouldn't list this slot in the form's required_slots caused training to fail. Now, this is only logged as a validation warning without causing the training to fail.

[3.12.7] - 2025-04-28

Rasa Pro 3.12.7 (2025-04-28)

Improvements

• Adds two optional properties on Jambonz channel connector, username and password which can be used to enable Basic Access Authentication

• Added support for basic authentication in Twilio channels (Voice Ready and Voice Streaming). This allows users to authenticate their Twilio channels using basic authentication credentials, enhancing security and access control for voice communication. To use this feature, set username and password in the Twilio channel configuration.

  credentials.yaml

  twilio_voice:
      username: your_username
      password: your_password
      ...
  
  twilio_media_streams:
      username: your_username
      password: your_password
      ...

  At Twilio, configure the webhook URL to include the basic authentication credentials:

  # twilio voice webhook
  https://<username>:<password>@yourdomain.com/webhooks/twilio_voice/webhook
  
  # twilio media streams webhook
  https://<username>:<password>@yourdomain.com/webhooks/twilio_media_streams/webhook

Bugfixes

• Fail rasa commands (rasa run, rasa inspect, rasa shell) when model file path doesn't exist instead of defaulting to the latest model file from the default directory /models.
• Fix the behaviour of action_hangup on Voice Inspector (browser_audio channel). Display that the session has ended
• Display a helpful error message in case of invalid API Key with Azure TTS
• Fixes Audiocodes Channel's event to intent mapping. All audiocode events are now mapped to an intent in the format vaig_event_<event>. That is, if Audiocodes sends an event noUserInput, the assistant will receive the intent /vaig_event_noUserInput

[3.12.6] - 2025-04-15

Rasa Pro 3.12.6 (2025-04-15)

Deprecations and Removals

• Remove the behaviour handling digressions as eligible flows that can be started while handling an active collect step. These properties have been removed from the flow and collect step:
  • ask_confirm_digressions
  • block_digressions Remove the new pattern pattern_handle_digressions.

Features

• Introduce a new boolean property force_slot_filling for the collect flow step. This property allows you to suppress incorrect predictions of the command generator or user digressions that are not relevant to the current slot filling. To enable this behavior, you should set the force_slot_filling property to True in the collect step of your flow configuration.

  flows:
      order_pizza:
        name: order pizza
        description: user asks for a pizza
        steps:
        - collect: pizza_type
        - collect: quantity
        - collect: address
          force_slot_filling: true
  When force_slot_filling is set to True, the command generator will only process the SetSlot command for the specified slot. By default, the property is set to False.

Bugfixes

• Security patch for Audiocodes and Genesys Channel connector Adds api_key (required) and client_secret (optional) properties to Genesys channel configuration Adds token (optional) property to Audiocodes-Stream channel
• Fix execution of custom validation action action_validate_slot_mappings by passing the extracted slot events to the tracker used when running this action.
• Make sure training always fail when domain is invalid.
• Add channel name to UserMessage created by the Audiocodes channel.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.12.5] - 2025-04-07

Rasa Pro 3.12.5 (2025-04-07)

Bugfixes

• Fix ChitChatAnswerCommand command replaced with CannotHandleCommand if there are no e2e stories defined. Improve validation that IntentlessPolicy has applicable responses: either responses in the domain that are not part of any flow, or if there are e2e stories. This validation performed during the training time and during cleanup of the ChitChatAnswerCommand command.
• • Fixes an issue with prompt rendering where minified JSON structures were displayed without properly escaping newlines, tabs, and quotes.
  • Introduced a new Jinja filter to_json_encoded_string that escapes newlines (\n), tabs (\t), and quotes (\") for safe JSON rendering. to_json_encoded_string filter preserves other special characters (e.g., umlauts) without encoding them.
  • Updated the default prompts for gpt-4o and claude-sonnet-3.5

[3.12.4] - 2025-04-01

Rasa Pro 3.12.4 (2025-04-01)

Bugfixes

• Send error event to the Kafka broker, when the original message size is too large, above the configured broker limit. This error handling mechanism was added to prevent Rasa-Pro server crashes.
• Update the following dependencies to versions that contain the latest patches for security vulnerabilities:
  • jinja2
  • werkzeug
  • cryptography
  • pyarrow
  • langchain
  • langchain-community
• Fix intermittent crashes on Inspector app when Enterprise Search or Chitchat Policies were triggered

[3.12.3] - 2025-03-26

Rasa Pro 3.12.3 (2025-03-26)

Bugfixes

• Fixes a bug in Inspector that raised a TypeError when serialising numpy.float64 from Tracker

[3.12.2] - 2025-03-25

Rasa Pro 3.12.2 (2025-03-25)

No significant changes.

[3.12.1] - 2025-03-21

Rasa Pro 3.12.1 (2025-03-21)

Bugfixes

• Fix filtering of StartFlow commands from LLM-based command generators during the overlap check with prior commands. When prior commands do not contain any StartFlow or HandleDigression commands, we should not filter out the StartFlow command from the LLM-based command generator.
• Remove:
  • cancel command when digression handling is defined
  • duplicate digression handling commands during command processing

Miscellaneous internal changes

Miscellaneous internal changes.

[3.12.0] - 2025-03-19

Rasa Pro 3.12.0 (2025-03-19)

Deprecations and Removals

• Deprecate MultiStepLLMCommandGenerator and schedule for removal in Rasa 4.0.0.
• Remove the beta feature flag check from the e2e testing with assertions feature. The RASA_PRO_BETA_E2E_ASSERTIONS environment variable is no longer needed as the feature is GA in 3.12.0.
• Deprecate the custom slot mapping type and its action slot mapping property which has been replaced with run_action_every_turn property name to retain backwards-compatible behavior.
• Deprecate the former list of dictionaries format for the condition key in a conditional response variation.

Features

• Add capability to use OAuth over Azure Entra ID for OpenAI instances deployed on Azure.

• Added Voice Stream Channel Connector for Genesys Cloud (AudioConnector Integration)

• Prevent unwanted digressions at collect flow steps by using one of the following new attributes available at both flow and collect step level:

  • ask_confirm_digressions: Asks the user to confirm if to continue with the current flow. Can be set to true or to a list of flow ids for which this behaviour should be activated.
  • block_digressions: Blocks any digression from the current flow and informs the user that they will return to the digression once the current flow is completed. Can be set to true or to a list of flow ids for which this behaviour should be activated.

  The above-mentioned behaviour is governed by a new pattern pattern_handle_digressions which is triggered only when the above attributes are used.

• Implement real-time validation of slot values.

  Add an optional property validation to slots to configure validations that should be run immediately. This property expects a list of rejections and a refill_utter which will be used to prompt users to provide a new value when validation fails.

  These validations are limited to common, reusable and universal checks that work independently of conversation context. For more complex validations that depend on conversation state, business logic, or external data, implement them in your flow definitions or custom actions instead.

• Introducing the CompactLLMCommandGenerator component, an enhancement over the SingleStepLLMCommandGenerator. This new component utilizes the highest-performing prompts for the models gpt-4o-2024-11-20 and claude-3-5-sonnet-20240620.

  To incorporate the CompactLLMCommandGenerator into your pipeline, simply add the following:

  pipeline:
    ...
    - name: CompactLLMCommandGenerator
    ...

• Multi-language support was implemented to enable the assistant to deliver localized responses and flow names that dynamically adjust to the user's language preference. In particular:

  • The default language is defined using the language key in config.yml, while additional supported languages are specified under additional_languages.
  • A translation section was introduced for responses to provide language-specific versions of the response text.
  • A translation section was added for flows to define localized flow names.
  • The rephraser prompt now accommodates the selected language.
  • Validation mechanisms were implemented to ensure proper use of translations; the CLI command rasa validate data translations is available for verification.
  • A new slot type, StrictCategoricalSlot, was developed to restrict its values to a predefined set.
  • A built-in language slot of the StrictCategoricalSlot type was added for managing translations effectively.

• [beta] Added Voice Stream channel connector to Audiocodes (audiocodes_stream)

Improvements

• Added validation to issue warnings for non-existent fixture/metadata names referenced in end-to-end tests.

• Implemented a fail fast mechanism that fails the end-to-end test case on the first failure, whether in user/bot turns or while using the assertions, to provide faster feedback.

• Added utterance_end_ms configuration to deepgram asr to handle noisy environments better

• Replace the optional dependency mlflow leveraged in the beta release of E2E testing with assertions when evaluating generative answers with custom prompts for each of the two generative metrics: generative_response_is_relevant and generative_response_is_grounded. This change now enables the usage of different LLM model providers and allows for a more flexible evaluation of generative components.

  Additionally, these generative assertions can make use of a new property utter_source (i.e. Enterprise Search, Contextual Rephraser or Intentless). This enables the assertion to be applied to a specific bot message source. It also for example prevents phrases such as Is there anything else i can help you with? triggered by pattern_completed to be checked for groundedness when the assertion should not be applied to it. Remove applying the same generative assertion to multiple bot messages in the same turn, however one bot message can be evaluated by multiple generative assertions in the same turn.

• Remove unnecessary deepcopy to improve performance in undo_fallback_prediction method of FallbackClassifier

• Add capability to control whether pattern_completed should execute when flow completes its execution. To control this behavior, a new parameter run_pattern_completed is added to the flow definition. By default this parameter is set to True which means pattern_completed will be executed when flow completes its execution (backward compatible). If this parameter is set to False, pattern_completed will not be executed when flow completes its execution.

• Allow slots to be filled by different slot extraction mechanisms (e.g. from_llm, predefined NLU-based mappings, custom actions etc.). Add new from_llm slot mapping boolean property allow_nlu_correction(by default set to False), which gives permission for LLM-issued SetSlot commands to correct slots previously filled via NLU-based mechanisms.

  Allow LLM-based command generators to issue other commands after NLUCommandAdapter has issued commands. Introduce a new LLM-based command generator config property minimize_num_calls (by default set to False) which maintains backwards compatibility with previous behaviour where LLM-based command generators were blocked from invoking the LLM after NLUCommandAdapter had issued commands.

  Update the default utterance utter_corrected_previous_input to use a new context property new_slot_values.

• Set the default priority for StartFlow commands issued by different command generator types i.e. NLUCommandAdapter or LLM-based command generator: When the different command generators issue StartFlow commands for different flows in the same user turn, the NLUCommandAdapter will always take priority while the LLM-based start flow command will be discarded.

  Remove the limitation that the NLUCommandAdapter must always precede the LLM-based command generator in the config pipeline.

• Introduce a new slot mapping type controlled that can be assigned to slots that are set via button payloads, set_slots flow steps or custom actions.

  Slots that solely use the new controlled slot mapping will not be available to be filled probabilistically by the NLU or LLM components. Note that this slot mapping can still be used alongside the other slot mapping types, however this comes with the risk of the slot being filled by the NLU or LLM components in a probabilistic manner.

• Slots with mappings of type controlled (formerly the now deprecated custom mapping type) can be set at every turn without its custom action having to be called explicitly by the user flow.

  If you are building a coexistence assistant where different controlled slots are set by custom actions in different subsystems, you must indicate which coexistence system is allowed to fill the slot. This is done by setting the coexistence_system property in the slot mapping configuration. This property is a string that must match one of the available categorical values: NLU, CALM, SHARED (when either system can set the slot).

• Support user to send a preformatted message to the invoke_llm method. This let's the user switch between the user and system roles when invoking the LLM model.

• Add support for pypred predicates in conditional response variations, similar to the usage of predicates in flows. The condition key in a response variation can now also be a string predicate that supports only the slots namespace. One of many logical operators supported is not.

• Make current slot type and its allowed values available for the prompt template rendering in SingleStepLLMCommandGenerator and CompactLLMCommandGenerator classes. Now you can use {{ current_slot_type }} and {{ current_slot_allowed_values }} placeholders in your custom prompt template.

• Made azure ASR endpoint and host, azure tts endpoint and cartesia tts endpoint configurable.

• Support usage of custom commands in the fine-tuning recipe.

• add support for mstts markups on azure TTS for improved SSML usage

Bugfixes

• Add the possibility to pass a transform callable parameter when writing yaml. This allows passing a custom function to transform endpoints before uploading to Studio. This was required to fix the issue where yaml wraps in quotes any string that doesn't start with an alphabetic character such as unexpanded environment variables in the endpoints yml file.
• Fixed the accuracy calculation to prevent 100% assertion reporting when a test case fails before any assertions are reached.
• Fixed regression on training time for projects with a lot of YAML files.
• Fix AvailableEndpoints to read from the default endpoints.yaml, if no endpoint is specified.
• Update domain yaml schema for conditional response condition type key to specify valid enum type as slot only.
• • Fixed an issue where the pattern_continue_interrupted was not correctly triggered when the flow digressed to a step containing a link.
• Add the flow ID as a prefix to step ID to ensure uniqueness. This resolves a rare bug where steps in a child flow with a structure similar to those in a parent flow (using a "call" step) could result in duplicate step IDs. In this case duplicates previously caused incorrect next step selection.
• Enable default action action_extract_slots to set slots that should be shared for coexistence in a NLU-based system, when the same slot can be requested and filled by a flow in the CALM system too.
• Fixed conversation stalling in AudioCodes channel by handling activities in background tasks. Previously, activities were processed synchronously which blocked responses to AudioCodes, causing request timeouts and activity retries. These retries would cancel ongoing processing and get rejected as duplicates. Now activities are processed asynchronously while responding immediately to AudioCodes requests.
• Improved error handling for Deepgram and Cartesia connection failures to display more meaningful error messages when authentication fails or other connection issues occur.
• Modify Enterprise Search Citation Prompt Template to use doc.text
• Fixes ClarifyCommand syntax in the fine-tuning recipe.
• Fixed a bug that lead to the response to silence timeouts being cut off
• Fixed a bug in Voice Inspector where the tracker (hence the conversation transcript) was only updated after the prediction loop was complete. The bug resulted in a perceived delay in case of slow custom actions where transcript was rendered after processing the complete conversation turn. Now the tracker is sent to the Inspector app after every iteration of prediction loop, which conveys a more accurate conversation state and transcript on the inspector app
• Fix passing the incorrect input type (user question text instead of bot answer text) to the prompt used by the generative_response_is_relevant assertion. Add instructions to both relevance and groundedness prompts to not add any more explanations to the LLM output apart from the expected json output to prevent parsing errors.
• Make real-time validation work with all slot types. Fixes bug where the same ValidateSlotPatternFlowStackFrame was being triggered multiple times.
• Handle multiple duplicate digressing flows occurring within the same flow:
  • if action_block_digressions runs for a found duplicate digressing flow already on the stack, it will not add it again
  • if action_continue_digressions runs for a found duplicate digressing flow already on the stack, it first removes it from the stack before pushing it to the top of the stack.
• Do not push the clarification pattern when the top user frame is an interruption frame.
• Consider linked and called flows as active flows when processing StartFlow commands.
• Fixed slot value injection in translated responses by updating response keys to interpolate.
• Updated language code parsing to enforce BCP 47 standard.
• Fix validation check that ensures that the slot used in the response condition is defined in the domain file.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.11.19] - 2025-08-19

Rasa Pro 3.11.19 (2025-08-19)

Bugfixes

• Fix correction of slots:

  • Slots can only be corrected in case they belong to any flow on the stack and the slot to be corrected is part of a collect step in any of those flows.
  • A correction of a slot should be applied if the flow that is about to start is using this slot.

• Don't tigger a slot correction for slots that are currently set to None as None counts as empty value.

• Fix issue where called flows could not set slots of their parent flow.

[3.11.18] - 2025-07-24

Rasa Pro 3.11.18 (2025-07-24)

Bugfixes

• Fix issues with bot not giving feedback for slot corrections.
• Fix validation of the FAISS documents folder to ensure it correctly discovers files in a recursive directory structure.
• Updated Inspector dependent packages (vite, and @adobe/css-tools) to address security vulnerabilities.
• Upgrade axios to fix security vulnerability.
• Upgrade litellm to fix security vulnerability.

[3.11.17] - 2025-07-03

Rasa Pro 3.11.17 (2025-07-03)

Bugfixes

• Flows now traverse called and linked flows, including nested and branching called / linked flows. As a result, E2E coverage reports include any linked and called flows triggered by the flow being tested.
• Fix issues where linked flows could not be cancelled and slots collected within linked flows could not be prefilled.
• Fix validation and improve error messages for the documents source directory when FAISS vector store is configured for the Enterprise Search Policy.
• Prevent slot correction when the slot is already set to the desired value.
• Fallback to CannotHandle command when the slot predicted by the LLM is not defined in the domain.
• Fix potential KeyError in EnterpriseSearchPolicy citation post-processing when the source does not have the correct citation. Improve the citation post-processing logic to handle additional edge cases.

[3.11.16] - 2025-06-12

Rasa Pro 3.11.16 (2025-06-12)

Bugfixes

• Ensure that old step ID formats (without the flow ID prefix) can be loaded without raising an InvalidFlowStepIdException in newer Rasa versions that expect the flow ID prefix.
• Fix an issue where running rasa inspect would always set the route_session_to_calm slot to True, even when no user-triggered commands were present. This caused incorrect routing to CALM, bypassing the logic of the router.
• Enable slot prefilling in patterns.

[3.11.15] - 2025-06-03

Rasa Pro 3.11.15 (2025-06-03)

Bugfixes

• Make domain an optional argument in CommandProcessorComponent. This change addresses a potential TypeError that could occur when loading a model trained without providing domain as a required argument. By making domain optional, models trained with older configurations or without a domain component will now load correctly without errors.

[3.11.14] - 2025-06-02

Rasa Pro 3.11.14 (2025-06-02)

Improvements

• • Updates the parameter name from max_tokens, which is deprecated by OpenAI, to max_completion_tokens. The old max_tokens is not supported for the o models.
  • Exposes LiteLLM's drop_params parameter for LLM configurations. :

Bugfixes

• The Clarify command now parses flow names with dashes.
• Add turn_wrapper to count multiple utterances by bot/user as single turn rather than individual turns. Always include last user utterance in rephraser prompt's conversation history.
• Remove StoryGraphProvider from the prediction graph in case IntentlessPolicy is not present to reduce the loading time of the bot in cases where the IntentlessPolicy is not used and a lot of stories are present.
• • Fixes default config initialization for the IntentlessPolicy.
• Fix remote model download when models are stored in a path, not in the root of the remote storage. Add new training CLI param --remote-root-only that can be used by the model service to store the model in the root of the remote storage. Propagate this parameter to the persistor's persist method. Simplify persistor code when retrieving models by downloading the model to the target path directly rather than copying the downloaded model to the target path. This also improved testability.
• When inspector is not used, root server path should output: Hello from Rasa: <version>.. When inspector is used, root server path should output HTML page with a link to the path on which inspector can be reached.

[3.11.13] - 2025-05-15

Rasa Pro 3.11.13 (2025-05-15)

Improvements

• Improved CRFEntityExtractor persistence and loading methods to improve model loading times.

Bugfixes

• Bumps aiohttp to 3.10.x, sentry-sdk to 2.8.x. Also bumps the locked versions for urllib3 and h11

[3.11.12] - 2025-05-14

Rasa Pro 3.11.12 (2025-05-14)

No significant changes.

[3.11.11] - 2025-05-08

Rasa Pro 3.11.11 (2025-05-08)

Bugfixes

• Fix issues in Audiocodes Channel Connector. The values in user_phone and bot_phone available in session_started_metadata are swapped to correctly map to calller and callee respectively. Fixed evening handling where events without the key "parameters" raised an exception.

[3.11.10] - 2025-05-06

Rasa Pro 3.11.10 (2025-05-06)

Bugfixes

• Implemented backtracking and corrected the recursion logic in the all-paths generation for a flow. Removed the need to deepcopy the step_ids_visited set on each branch within _handle_links, which prevents the coverage report from freezing due to hitting the recursion limit.

[3.11.9] - 2025-04-30

Rasa Pro 3.11.9 (2025-04-30)

Bugfixes

• Upgrade openai and litellm dependencies to fix found vulnerabilities in litellm.
• Add support for auth_token to Rasa Inspector.
• Fixed issue where a slot mapping referencing a form in the active_loop slot mapping conditions that wouldn't list this slot in the form's required_slots caused training to fail. Now, this is only logged as a validation warning without causing the training to fail.

[3.11.8] - 2025-04-28

Rasa Pro 3.11.8 (2025-04-28)

Improvements

• Adds two optional properties on Jambonz channel connector, username and password which can be used to enable Basic Access Authentication

• Added support for basic authentication in Twilio channels (Voice Ready and Voice Streaming). This allows users to authenticate their Twilio channels using basic authentication credentials, enhancing security and access control for voice communication. To use this feature, set username and password in the Twilio channel configuration.

  credentials.yaml

  twilio_voice:
      username: your_username
      password: your_password
      ...
  
  twilio_media_streams:
      username: your_username
      password: your_password
      ...

  At Twilio, configure the webhook URL to include the basic authentication credentials:

  # twilio voice webhook
  https://<username>:<password>@yourdomain.com/webhooks/twilio_voice/webhook
  
  # twilio media streams webhook
  https://<username>:<password>@yourdomain.com/webhooks/twilio_media_streams/webhook

Bugfixes

• Fixed a bug that lead to the response to silence timeouts being cut off
• Fail rasa commands (rasa run, rasa inspect, rasa shell) when model file path doesn't exist instead of defaulting to the latest model file from the default directory /models.
• Fix the behaviour of action_hangup on Voice Inspector (browser_audio channel). Display that the session has ended
• Display a helpful error message in case of invalid API Key with Azure TTS
• Fixes Audiocodes Channel's event to intent mapping. All audiocode events are now mapped to an intent in the format vaig_event_<event>. That is, if Audiocodes sends an event noUserInput, the assistant will receive the intent /vaig_event_noUserInput

[3.11.7] - 2025-04-14

Rasa Pro 3.11.7 (2025-04-14)

Bugfixes

• Fix ChitChatAnswerCommand command replaced with CannotHandleCommand if there are no e2e stories defined. Improve validation that IntentlessPolicy has applicable responses: either responses in the domain that are not part of any flow, or if there are e2e stories. This validation performed during the training time and during cleanup of the ChitChatAnswerCommand command.
• Security patch for Audiocodes channel connector. Audiocodes channel was only checking for the existence of authentication token. It now does a constant-time string comparison of the authentication token in connection request with that provided in channel configruation.
• Make sure training always fail when domain is invalid.
• Add channel name to UserMessage created by the Audiocodes channel.

[3.11.6] - 2025-04-02

Rasa Pro 3.11.6 (2025-04-02)

Bugfixes

• Improved error handling for Deepgram and Cartesia connection failures to display more meaningful error messages when authentication fails or other connection issues occur.
• Modify Enterprise Search Citation Prompt Template to use doc.text
• Fixes ClarifyCommand syntax in the fine-tuning recipe.
• Send error event to the Kafka broker, when the original message size is too large, above the configured broker limit. This error handling mechanism was added to prevent Rasa-Pro server crashes.
• Update the following dependencies to versions that contain the latest patches for security vulnerabilities:
  • jinja2
  • werkzeug
  • requests
  • cryptography
  • pyarrow
  • langchain
  • langchain-community

[3.11.5] - 2025-02-18

Rasa Pro 3.11.5 (2025-02-18)

Bugfixes

• Updated Inspector dependent packages (cross-spawn, mermaid, dom-purify, vite, braces, ws, axios and rollup) to address security vulnerabilities.
• Enable default action action_extract_slots to set slots that should be shared for coexistence in a NLU-based system, when the same slot can be requested and filled by a flow in the CALM system too.
• Fixed conversation stalling in AudioCodes channel by handling activities in background tasks. Previously, activities were processed synchronously which blocked responses to AudioCodes, causing request timeouts and activity retries. These retries would cancel ongoing processing and get rejected as duplicates. Now activities are processed asynchronously while responding immediately to AudioCodes requests.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.11.4] - 2025-01-30

Rasa Pro 3.11.4 (2025-01-30)

Improvements

• Remove unnecessary deepcopy to improve performance in undo_fallback_prediction method of FallbackClassifier

Bugfixes

• • Fixed an issue where the pattern_continue_interrupted was not correctly triggered when the flow digressed to a step containing a link.
• Add the flow ID as a prefix to step ID to ensure uniqueness. This resolves a rare bug where steps in a child flow with a structure similar to those in a parent flow (using a "call" step) could result in duplicate step IDs. In this case duplicates previously caused incorrect next step selection.
• Optimized the DirectCustomActionExecutor by registering custom actions only once.
• Updated cryptography and anyio to resolve security vulnerabilities.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.11.3] - 2025-01-14

Rasa Pro 3.11.3 (2025-01-14)

Improvements

• Enhances YAML parser to validate environment variable resolution for sensitive keys.

Bugfixes

• Add flow yaml validation when using the HTTP API /model/train endpoint. An invalid flow yaml will return a 400 response status code with a message describing the error.
• Make pattern_session_start work with rasa inspector to allow the assistant proactively start the conversation with a user.
• Fix writing the test cases obtained via the e2e test case conversion command to file, where test_cases key was written as a list item, instead of a dict key. This caused running the test cases to fail because it didn't comply with the e2e test schema. This PR fixes the issue by writing the test cases as a dict key.
• Fixed Inspector's Tracker State view not updating in real-time by moving story fetch logic into WebSocket message handler. Previously, story updates were only triggered on session ID changes, causing stale tracker state after the first conversation turn.
• Add pre-training custom validation to the domain responses that would raise a Rasa Pro validation error when a domain response is an empty sequence.
• Fixes a critical security vulnerability with jsonpickle dependency by upgrading to the patched version.
• Updated pymilvus and minio to address security vulnerability.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.11.2] - 2024-12-19

Rasa Pro 3.11.2 (2024-12-19)

Bugfixes

• Validate that api_type key is only used for supported providers (Azure and OpenAI).
• Enable asserting events returned by action_session_start when running end-to-end testing with assertions format. The following assertions can be used:
  • slot_was_set
  • slot_was_not_set
  • bot_uttered
  • bot_did_not_utter
  • action_executed
• Fixed voice inspector to work with any URL by dynamically constructing WebSocket URL from current domain. This enables voice testing in GitHub Codespaces and other remote environments.
• • Fixed an error in rasa llm finetune prepare-data when using a subclass of SingleStepLLMCommandGenerator.
  • Resolved an issue where rasa llm finetune prepare-data did not support model groups.
• Fix AvailableEndpoints to read from the default endpoints.yaml, if no endpoint is specified.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.11.1] - 2024-12-13

Rasa Pro 3.11.1 (2024-12-13)

Bugfixes

• Add the possibility to pass a transform callable parameter when writing yaml. This allows passing a custom function to transform endpoints before uploading to Studio. This was required to fix the issue where yaml wraps in quotes any string that doesn't start with an alphabetic character such as unexpanded environment variables in the endpoints yml file.
• Pass flow human-readable name instead of flow id when the cancel pattern stack frame is pushed during flow policy validation checks of collect steps.
• Fixed the accuracy calculation to prevent 100% assertion reporting when a test case fails before any assertions are reached.
• Fixed regression on training time for projects with a lot of YAML files.

[3.11.0] - 2024-12-11

Rasa Pro 3.11.0 (2024-12-11)

Deprecations and Removals

• Removed UnexpecTEDIntentPolicy from the default config.yml. It is an experimental policy and not suitable for default configuration
• The reset_after_flow_ends property of collect steps is now deprecated and will be removed in Rasa Pro 4.0.0. Please use the persisted_slots property at the flow level instead.

Features

• Added Twilio Media Streams channel which can be configured to use arbitrary Text-To-Speech and Speech-To-Text services. Added Voice Stream Channel Interface which makes it easier to add voice channels that directly integrate with audio streams. Added support for Deepgram Speech-To-Text and Azure Text-To-Speech in Voice Stream Channels.

• Added default action action_hangup it can be used to hang up a phone call from a flow. Added SessionEnded event and SessionEndCommand command Updated Audiocodes, Jambonz and Twilio Voice channels to send /session_end if the phone call is disconnected by user.

• Added support for Cartesia Text-To-Speech in Voice Stream Channels.

• Implement Rasa Pro native model service that takes care of training and running an assistant model in Studio. To find out more about this service, read more in the Studio documentation.

• Added a feature to be able to use voice to interact with the bot in the inspector.

• Multi-LLM Routing:

  1. Decoupled LLM Configuration from Components

     • The previous integration of LLMs within CALM is closely tied to the components where they are used. However, this is no longer necessary, as we no longer perform training within the individual components that interact with external LLM endpoints.
     • As a result, LLM and embedding client configurations have been moved to endpoints.yml. To define LLM configurations in endpoints.yml, use the model_groups as shown below:

       model_groups:
         - id: gpt-4-direct
           models:
             - provider: openai
               model: gpt-4
               timeout: 7
               temperature: 0.0
         - id: text-embedding-3-small-direct
           models:
             - provider: openai
               model: text-embedding-3-small
     • These model_groups can then be referenced in config.yml as follows:

       pipeline:
         ...
         - name: SingleStepLLMCommandGenerator
           llm:
             model_group: gpt-4-direct
           flow_retrieval:
             embeddings:
               model_group: text-embedding-3-smal-direct
         ...

  2. Support for Multiple Subscription Deployments

     • Allows customers to use deployments from different subscriptions for the same provider.
     • Resolved the limitation of API key configuration being tied exclusively to a single environment variable.

     Example configuration in endpoints.yml for Azure deployments:

     model_groups:
       - id: azure-gpt-model-eu
         models:
            - provider: azure
              deployment: azure-eu-deployment
              api_base: https://api.azure-europe.example.com
              api_version: 2024-08-01-preview
              api_key: ${AZURE_API_KEY_EU}
              timeout: 7
              temperature: 0.0
              ...
       - id: azure-gpt-model-us
         models:
            - provider: azure
              deployment: azure-us-deployment
              api_base: https://api.azure-us.example.com
              api_version: 2024-08-01-preview
              api_key: ${AZURE_API_KEY_US}
              timeout: 7
              temperature: 0.0
              ...
          ...

  3. Seamless Model Configuration Across Environments Without Retraining

     • Added support for using different model configurations in different environments, such as dev, staging, and prod, without requiring the bot to be retrained for each environment.
     • Extended the ${...} syntax to deployment, api_base, and api_version in model_groups, allowing these values to change dynamically based on the environment.

     model_groups:
     - id: azure-gpt-4
       models:
         - provider: azure
           deployment: ${AZURE_DEPLOYMENT_GPT4}
           api_base: ${AZURE_API_BASE_GPT4}
           api_key: ${AZURE_API_KEY_GPT4}
           ...
     - id: azure-text-embeddings-3-small
       models:
         - provider: azure
           deployment: ${AZURE_DEPLOYMENT_EMBEDDINGS_3_SMALL}
           api_base: ${AZURE_API_BASE_EMBEDDINGS_3_SMALL}
           api_key: ${AZURE_API_EMBEDDINGS_3_SMALL}
           ...

  4. Supporting Multiple Deployments for Load Balancing

     • Enabled targeting of multiple LLM deployments for a single Rasa component.
     • Implemented the routing feature that supports load balancing to handle rate limits and improve scalability. When multiple models are defined within a model group, you can specify the router key with a routing_strategy to control how requests are distributed among the models.

     Example configuration in endpoints.yml for Azure deployments with load balancing:

     model_groups:
       - id: azure-gpt-models
          models:
            - provider: azure
              deployment: azure-eu-deployment
              api_base: https://api.azure-europe.example.com
              api_version: 2024-08-01-preview
              api_key: ${AZURE_API_KEY_EU}
              timeout: 7
              temperature: 0.0
              ...
            - provider: azure
              deployment: azure-us-deployment
              api_base: https://api.azure-us.example.com
              api_version: 2024-08-01-preview
              api_key: ${AZURE_API_KEY_US}
              timeout: 7
              temperature: 0.0
              ...
          router:
            routing_strategy: least-busy
          ...

     Example of usage in config.yml:

     pipeline:
     ...
       - name: SingleStepLLMCommandGenerator
         llm:
           model_group: azure-gpt-models
           ...

  5. Backward Compatibility

     • Existing configurations that couple LLMs to specific Rasa components remain unaffected by this change.
     • However, this configuration method is now deprecated and scheduled for removal in version 4.0.0.

• Added support for Azure Speech-To-Text in Voice Stream Channels.

• Added UserSilenceCommand and pattern_user_silence which is triggered by Voice Stream channels when the user is silent for more than a silence timeout. These values are configurable with the newly added slots silence_timeout and consecutive_silence_timeouts. Silence Monitoring is disabled by default and can be enabled using the configuration monitor_silence: true in the relevant Voice Stream Channel configuration.

• The inspector is not its own input / output channel anymore. Rather, it can be attached to other channels. This way, it isn't limited to conversations going through the socketio channel anymore, but can be used with other text channels or voice channels.

  You can attach it to any channel(s) configured in your credentials.yml by adding a flag to rasa run: rasa run --inspect.

  In addition to that, the conenience cli command rasa inspect is retained, which starts the inspector with the socketio channel as usual.

Improvements

• In Audiocodes channel, /vaig_event_start is replaced by /session_start. This intent marks the beginning of conversation and it is sent when the phone call is connected.

• Introduced the environment variable MAX_NUMBER_OF_PREDICTIONS_CALM to configure the CALM-specific limit for the number of predictions. This variable defaults to 1000, providing a higher prediction limit compared to the default value of 10 for nlu-based assistants.

• In Audiocodes and Twilio Voice channel connector, the call metadata received from the providers can be accessed in the slot session_started_metadata. The call metadata parameter names have been standardised with CallParameters dataclass Twilio Voice Channel Connector sends /session_start intent at the beginning of conversation and the channel parameter initial_prompt has been removed

• Enable configurability of Vault secret manager's mount point property in the endpoints yaml file or as an environment variable.

• In Twilio Media Streams channel connector, call metadata is availble in session_start_metadata slot. It also supports default action action_hangup

• Catch API connection errors, and validate the correctness of the values present in model configuration at model training time by making a test API request. This feature is enabled by default and can be disabled by setting the environment variable LLM_API_HEALTH_CHECK to False.

• Socketio channel connector now sends the websocket messages tracker_state and rasa_events with each bot response. tracker_state contains the tracker store state at that point in conversation and includes slots, events, stack, latest message and latest action. rasa_events contains a list of new events that have happened since the last message.

• Speech-To-Text and Text-To-Speech Services can be configured for Voice Stream Channel Connectors Added tests for voice components and redefined code structure

• Add support for Python 3.11

• Removed JSON response validation except when HTTP protocol and E2E Stub is used for Custom Action execution.

• Optimized JSON response validation by initializing the Draft202012Validator once and caching it.

• Add an optional property persisted_slots at the flow level. This property configures whether slots collected or set across any of the flow steps should be persisted after the flow ends. This property expects a list of slot names.

• Added support for custom Automatic Speech Recognition (ASR) or Text To Speech (TTS) providers to a Rasa Assistant. This allows developers to bring their own speech providers to Rasa by subclassing classes ASREngine and TTSEngine

• If flow retrieval is disabled, a warning is raised only if the number of user flows exceed 20.

• Added validation to the TestCase class to issue a warning when duplicate user messages lack metadata or have incorrect metadata. This enhancement provides clear guidance to users on the issue and how to resolve it.

• Fixed global should-hangup variable in Voice Stream Channels by moving to a context variable CallState that stores the session variables

• Run Rasa Pro data validation before uploading to Studio. This is to avoid uploading invalid assistant data that would raise errors during Rasa Pro model training in Studio.

• Added vector_name to Qdrant's configuration to enable customization of the vector field name for storing embeddings.

• Enhanced YamlValidationException error messages to include the line number and a relevant YAML snippet showing where the validation error occurred. Line numbers start from 1 (1-based indexing).

  The error-handling behavior has been modified so that only one validation error is displayed. This exception is raised when the YAML content does not comply with the defined YAML schema.

• Added a new assertion type bot_did_not_utter to allow testing that the bot does not utter specific messages or include certain buttons during conversations.

• Ensure that the model service fails properly if the minimum disk space requirement is not met.

• Do not expand environment variables when reading yaml files during rasa studio upload execution.

• Stream model files to Studio rather than providing full files. Provide a HEAD endpoint for Studio to check if a model is available and what its size is. Add an environment variable to set the port of the model service. This makes the development with Studio easier, previously the port was hard coded making it harder to use a separately deployed model service now that Studio includes that in its development deployment.

• Add flag --skip-yaml-validation to skip YAML validation during Rasa run. User can use it to skip domain YAML validation during Rasa run. Do not instantiate multiple instances of TrainingDataImporter class for validation and training.

• Introduced a summarize_history flag for the contextual response rephraser, defaulting to True. When set to False, the conversation transcript instead of the summary is included in the prompt of the contextual response rephraser. This saves a separate summarization call to an LLM. The number of conversation turns to be used when summarize_history is set to False can be set via max_historical_turns. By default this value is set to 5.

  Example:

  nlg:
    - type: rephrase
      summarize_history: False
      max_historical_turns: 5

Bugfixes

• Fix OpenAI LLM client ignoring API base and API version arguments if set.

• Fix AttributeError with the instrumentation of the run method of the CustomActionExecutor class.

• Throw DuplicatedFlowIdException during rasa data validate and rasa train if there are duplicate flows defined.

• Replace pickle and joblib with safer alternatives, e.g. json, safetensors, and skops, for serializing components.

  Note: This is a model breaking change. Please retrain your model.

  If you have a custom component that inherits from one of the components listed below and modified the persist or load method, make sure to update your code. Please contact us in case you encounter any problems.

  Affected components:

  • CountVectorFeaturizer
  • LexicalSyntacticFeaturizer
  • LogisticRegressionClassifier
  • SklearnIntentClassifier
  • DIETClassifier
  • CRFEntityExtractor
  • TrackerFeaturizer
  • TEDPolicy
  • UnexpectedIntentTEDPolicy

• Avoid filling slots that have ask_before_filling = True and utilize a from_text slot mapping during other steps in the flow. Ensure that the NLUCommandAdapter only fills these types of slots when the flow reaches the designated collection step.

• Check for the metadata's step_id and active_flow keys when adding the ActionExecuted event to the flows paths stack.

• Fixed a bug on Windows where flow files with names starting with 'u' would fail to load due to improper path escaping in YAML content processing

• Fixes OpenAIException - AsyncClient.init() got an unexpected keyword argument 'proxies'

• Fix retrieval of model file stored in the cloud storage by the model service. This change consisted in uploading only the model file instead of the full model path during training when --remote-storage CLI flag is used.

• Fix issue in e2e testing when customising action_session_start would lead to AttributeError, because the output_channel was not set. This is now fixed by setting the output_channel to CollectingOutputChannel().

Miscellaneous internal changes

Miscellaneous internal changes.

[3.10.27] - 2025-06-03

Rasa Pro 3.10.27 (2025-06-03)

Bugfixes

• Make domain an optional argument in CommandProcessorComponent. This change addresses a potential TypeError that could occur when loading a model trained without providing domain as a required argument. By making domain optional, models trained with older configurations or without a domain component will now load correctly without errors.

[3.10.26] - 2025-06-02

Rasa Pro 3.10.26 (2025-06-02)

Improvements

• • Updates the parameter name from max_tokens, which is deprecated by OpenAI, to max_completion_tokens. The old max_tokens is not supported for the o models.
  • Exposes LiteLLM's drop_params parameter for LLM configurations.

Bugfixes

• The Clarify command now parses flow names with dashes.
• Add turn_wrapper to count multiple utterances by bot/user as single turn rather than individual turns. Always include last user utterance in rephraser prompt's conversation history.
• Remove StoryGraphProvider from the prediction graph in case IntentlessPolicy is not present to reduce the loading time of the bot in cases where the IntentlessPolicy is not used and a lot of stories are present.
• • Fixes default config initialization for the IntentlessPolicy.

[3.10.25] - 2025-05-15

Rasa Pro 3.10.25 (2025-05-15)

Improvements

• Improved CRFEntityExtractor persistence and loading methods to improve model loading times.

Bugfixes

• Bumps aiohttp to 3.10.x, sentry-sdk to 2.8.x. Also bumps the locked versions for urllib3 and h11

[3.10.24] - 2025-05-14

Rasa Pro 3.10.24 (2025-05-14)

No significant changes.

[3.10.23] - 2025-05-06

Rasa Pro 3.10.23 (2025-05-06)

Bugfixes

• Implemented backtracking and corrected the recursion logic in the all-paths generation for a flow. Removed the need to deepcopy the step_ids_visited set on each branch within _handle_links, which prevents the coverage report from freezing due to hitting the recursion limit.

[3.10.22] - 2025-04-30

Rasa Pro 3.10.22 (2025-04-30)

Bugfixes

• Add support for auth_token to Rasa Inspector.

[3.10.21] - 2025-04-29

Rasa Pro 3.10.21 (2025-04-29)

Bugfixes

• Fixed issue where a slot mapping referencing a form in the active_loop slot mapping conditions that wouldn't list this slot in the form's required_slots caused training to fail. Now, this is only logged as a validation warning without causing the training to fail.

[3.10.20] - 2025-04-28

Rasa Pro 3.10.20 (2025-04-28)

Improvements

• Added support for basic authentication in Twilio voice channel. This allows users to authenticate their Twilio voice channel using basic authentication credentials, enhancing security and access control for voice communication. To use this feature, set username and password in the Twilio channel configuration.

  credentials.yaml

  twilio_voice:
      username: your_username
      password: your_password
      ...

  At Twilio, configure the webhook URL to include the basic authentication credentials:

  # twilio voice webhook
  https://<username>:<password>@yourdomain.com/webhooks/twilio_voice/webhook

Bugfixes

• Fail rasa commands (rasa run, rasa inspect, rasa shell) when model file path doesn't exist instead of defaulting to the latest model file from the default directory /models.
• Upgrade openai and litellm dependencies to fix found vulnerabilities in litellm.

[3.10.19] - 2025-04-15

Rasa Pro 3.10.19 (2025-04-15)

Bugfixes

• Fix ChitChatAnswerCommand command replaced with CannotHandleCommand if there are no e2e stories defined. Improve validation that IntentlessPolicy has applicable responses: either responses in the domain that are not part of any flow, or if there are e2e stories. This validation performed during the training time and during cleanup of the ChitChatAnswerCommand command.
• Security patch for Audiocodes channel connector. Audiocodes channel was only checking for the existence of authentication token. It now does a constant-time string comparison of the authentication token in connection request with that provided in channel configruation.
• Make sure training always fail when domain is invalid.
• Add channel name to UserMessage created by the Audiocodes channel.

[3.10.18] - 2025-04-02

Rasa Pro 3.10.18 (2025-04-02)

Bugfixes

• Updated Inspector dependent packages (cross-spawn, mermaid, dom-purify, vite, braces, ws, axios and rollup) to address security vulnerabilities.
• Modify Enterprise Search Citation Prompt Template to use doc.text
• Fixes ClarifyCommand syntax in the fine-tuning recipe.
• Send error event to the Kafka broker, when the original message size is too large, above the configured broker limit. This error handling mechanism was added to prevent Rasa-Pro server crashes.
• Update the following dependencies to versions that contain the latest patches for security vulnerabilities:
  • jinja2
  • werkzeug
  • requests
  • cryptography
  • pyarrow
  • langchain
  • langchain-community

[3.10.17] - 2025-01-30

Rasa Pro 3.10.17 (2025-01-30)

Improvements

• Remove unnecessary deepcopy to improve performance in undo_fallback_prediction method of FallbackClassifier

Bugfixes

• • Fixed an issue where the pattern_continue_interrupted was not correctly triggered when the flow digressed to a step containing a link.
• Add the flow ID as a prefix to step ID to ensure uniqueness. This resolves a rare bug where steps in a child flow with a structure similar to those in a parent flow (using a "call" step) could result in duplicate step IDs. In this case duplicates previously caused incorrect next step selection.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.10.16] - 2025-01-15

Rasa Pro 3.10.16 (2025-01-15)

Bugfixes

• • Fixed an error in rasa llm finetune prepare-data when using a subclass of SingleStepLLMCommandGenerator.
• Make pattern_session_start work with rasa inspector to allow the assistant proactively start the conversation with a user.
• Fix writing the test cases obtained via the e2e test case conversion command to file, where test_cases key was written as a list item, instead of a dict key. This caused running the test cases to fail because it didn't comply with the e2e test schema. This PR fixes the issue by writing the test cases as a dict key.
• Add pre-training custom validation to the domain responses that would raise a Rasa Pro validation error when a domain response is an empty sequence.
• Fixes a critical security vulnerability with jsonpickle dependency by upgrading to the patched version.
• Updated pymilvus and minio to address security vulnerability.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.10.15] - 2024-12-18

Rasa Pro 3.10.15 (2024-12-18)

Bugfixes

• Validate that api_type key is only used for supported providers (Azure and OpenAI).
• Fix issue in e2e testing when customising action_session_start would lead to AttributeError, because the output_channel was not set. This is now fixed by setting the output_channel to CollectingOutputChannel().
• Fixed the accuracy calculation to prevent 100% assertion reporting when a test case fails before any assertions are reached.
• Pass flow human-readable name instead of flow id when the cancel pattern stack frame is pushed during flow policy validation checks of collect steps.
• Try to instantiate LLM/embeddings client when loading component to validate environment variables.
• Enable asserting events returned by action_session_start when running end-to-end testing with assertions format. The following assertions can be used:
  • slot_was_set
  • slot_was_not_set
  • bot_uttered
  • action_executed

[3.10.14] - 2024-12-04

Rasa Pro 3.10.14 (2024-12-04)

Bugfixes

• Avoid filling slots that have ask_before_filling = True and utilize a from_text slot mapping during other steps in the flow. Ensure that the NLUCommandAdapter only fills these types of slots when the flow reaches the designated collection step.
• Fixes OpenAIException - AsyncClient.init() got an unexpected keyword argument 'proxies'
• Fix validation for LLM/Embedding clients when the api_base is configured in the config itself but not as an environment variable.

[3.10.13] - 2024-11-29

Rasa Pro 3.10.13 (2024-11-29)

Bugfixes

• Implement eq and hash functions for ChangeFlowCommand to fix error=unhashable type: 'ChangeFlowCommand' error in MultiStepCommandGenerator.
• Fixed an issue on Windows where flow files with names starting with 'u' would fail to load due to improper path escaping in YAML content processing
• Store the value of the --disable-verify CLI flag in the disable_verify attribute of the StudioConfig object, so it can be reused across other studio commands.

[3.10.12] - 2024-11-25

Rasa Pro 3.10.12 (2024-11-25)

Bugfixes

• Replace pickle and joblib with safer alternatives, e.g. json, safetensors, and skops, for serializing components.

  Note: This is a model breaking change. Please retrain your model.

  If you have a custom component that inherits from one of the components listed below and modified the persist or load method, make sure to update your code. Please contact us in case you encounter any problems.

  Affected components:

  • CountVectorFeaturizer
  • LexicalSyntacticFeaturizer
  • LogisticRegressionClassifier
  • SklearnIntentClassifier
  • DIETClassifier
  • CRFEntityExtractor
  • TrackerFeaturizer
  • TEDPolicy
  • UnexpectedIntentTEDPolicy

[3.10.11] - 2024-11-20

Rasa Pro 3.10.11 (2024-11-20)

Bugfixes

• Fix parsing of commands in case the LLM response surrounds flow names, slot names, or slot values with single or double quotes.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.10.10] - 2024-11-14

Rasa Pro 3.10.10 (2024-11-14)

Bugfixes

• Check for the metadata's step_id and active_flow keys when adding the ActionExecuted event to the flows paths stack.

[3.10.9] - 2024-11-13

Rasa Pro 3.10.9 (2024-11-13)

Bugfixes

• Introduced the environment variable MAX_NUMBER_OF_PREDICTIONS_CALM to configure the CALM-specific limit for the number of predictions. This variable defaults to 1000, providing a higher prediction limit compared to the default value of 10 for nlu-based assistants.
• Filter out comments from e2e test input files when writing e2e results to file.
• Specified UTF-8 encoding to correctly read test cases on Windows.

[3.10.8] - 2024-10-24

Rasa Pro 3.10.8 (2024-10-24)

Bugfixes

• The user message "/restart" is now restarting the session again after adding a proper implementation (stack frame and command) for pattern_restart.
• Only infer and set the provider to azure for our LLM clients in case NO provider is specified, but the deployment key is set.
• Fix OPENAI_API_KEY authentication error when using self-hosted provider.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.10.7] - 2024-10-17

Rasa Pro 3.10.7 (2024-10-17)

Improvements

• Change default response of utter_free_chitchat_response from "placeholder_this_utterance_needs_the_rephraser" to "Sorry, I'm not able to answer that right now.".

Bugfixes

• Disallow using the command payload syntax to set slots not filled by any of the active or startable flow(s) collect steps.
• Add flow name to error message validator.verify_flows_steps_against_domain.collect_step.
• Update e2e test results output files on each test run so that, for example, when all tests pass on subsequent runs after failing previously, the failed results output file is emptied.
• Disable strict SSL verification to the Rasa Studio authentication server via the --disable-verify or -x CLI argument added to the rasa studio config command.
• Upgrade zipp dependency version to fix a security vulnerability: CVE-2024-5569.

[3.10.6] - 2024-10-04

Rasa Pro 3.10.6 (2024-10-04)

Bugfixes

• Fix cleanup of SetSlot commands issued by the LLM-based command generator for slots that define a slot mapping other than the from_llm slot mapping. The command processor now correctly removes the SetSlot command in these scenarios and instead adds a CannotHandleCommand.
• Fix UnicodeDecodeError while reading Windows path from yaml files.
• Fix model loading from remote storage by correcting the handling of remote storage enum during the creation of the persistor object.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.10.5] - 2024-10-01

Rasa Pro 3.10.5 (2024-10-01)

Bugfixes

• Fix the case where IntentlessPolicy is triggered while no e2e stories were written to guide it. In this situation a CannotHandleCommand will be issued.

• Update litellm to version 1.45.0 to fix security vulnerability (CVE-2024-6587). Update gitpython to version 3.1.41 to fix security vulnerability (CVE-2024-22190). Update certifi to version 2024.07.04 to fix security vulnerability (CVE-2024-39689).

• Prevent invalid domain with incorrectly defined intent from throwing stack trace. Throw InvalidDomain exception and send message to the user instead. The message looks like this:

  Detected invalid intent definition: {'intent': 'ask_help'}.
  Please make sure all intent definitions are valid.

• Support text completions endpoint when using self hosted models.

  The use_chat_completions_endpoint parameter is now supported when using self-hosted models. This parameter is used to enable the use of the chat completions endpoint when using a self-hosted model. This parameter is set to True by default. To use the text completions endpoint, set use_chat_completions_endpoint to False in the llm section of the component.

  Usage:

  llm:
      provider: self-hosted
      model: meta-llama/Meta-Llama-3-8B
      api_base: "https://my-endpoint/v1"
      use_chat_completions_endpoint: false

• Fixes an issue where the CountVectorsFeaturizer and LogisticRegressionClassifier would throw error during inference when no NLU training data is provided.

• Added tracing explicitly to GRPCCustomActionExecutor.run in order to pass the tracing context to the action server.

[3.10.4] - 2024-09-25

Rasa Pro 3.10.4 (2024-09-25)

Bugfixes

• Fix failing validation of categorical slots when slot values contain Apostrophe.

[3.10.3] - 2024-09-20

Rasa Pro 3.10.3 (2024-09-20)

No significant changes.

[3.10.2] - 2024-09-19

Rasa Pro 3.10.2 (2024-09-19)

Deprecations and Removals

• Dropped support for Python 3.8 ahead of Python 3.8 End of Life in October 2024. In Rasa Pro versions 3.10.0, 3.9.11 and 3.8.13, we needed to pin the TensorFlow library version to 2.13.0rc1 in order to remove critical vulnerabilities; this resulted in poor user experience when installing these versions of Rasa Pro with uv pip. Removing support for Python 3.8 will make it possible to upgrade to a stabler version of TensorFlow.

Improvements

• Update Keras and Tensorflow to version 2.14. This will eliminate the need to use the --prerelease allow flag when installing Rasa Pro using uv pip tool.

Bugfixes

• Revert the old behavior when loading trained model by supplying a path to the model on the remote storage by using the model path (-m) argument when REMOTE_STORAGE_PATH environment variable is not set. Resulting path on the remote storage will be the same as the model path (-m) argument.

  Additionally, entire model path (-m) argument wil be used when trained model is being uploaded to the remote storage with REMOTE_STORAGE_PATH environment variable not set. Resulting path on the remote storage will be the same as the model path (-m) argument.

  If REMOTE_STORAGE_PATH environment variable is set, only the file name part of the model path (-m) argument is used in both loading and storage from/to the remote storage. Resulting path on the remote storage will be: REMOTE_STORAGE_PATH + file name part of the model path (-m) argument.

• Fixed UnexpecTEDIntentlessPolicy training errors that resulted from a change to batching behavior. Changed the batching behavior back to the original for all components. Made the changed batching behavior accessible in DietClassifier using drop_small_last_batch: True.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.10.1] - 2024-09-11

Rasa Pro 3.10.1 (2024-09-11)

Bugfixes

• Fix OpenAI LLM client ignoring API base and API version arguments if set.
• Fix FileNotFound error when running rasa studio commands and no pre-existing local assistant project exists.
• Fixed telemetry collection for the components Rephraser, LLM Intent Classifier, Intentless Policy and Enterprise Search Policy to ensure that the telemetry data is only collected when it is enabled
• Update the default config for E2E test conversion to use the provider key instead of api_type.
• Fix inconsistent recording of telemetry events for llm-based command generators.
• Throw deprecation warning when REQUESTS_CA_BUNDLE env var is used.

[3.10.0] - 2024-09-04

Rasa Pro 3.10.0 (2024-09-04)

Deprecations and Removals

• Remove experimental LLMIntentClassifier. Use Rasa CALM instead.

Features

• Implement the shell output of accuracy rate by assertion type as a table when running end-to-end testing with assertions.

• Implement E2E testing assertions that measure metrics such as grounded-ness and answer relevance of generative responses issued by either Enterprise Search or the Contextual Response Rephraser.

  You must specify a threshold which must be reached for the generative evaluation assertion to pass. In addition, you can also specify ground_truth if you prefer providing this in the E2E test rather than relying on the retrieved context from the vector store (in the case of Enterprise Search) or from the domain (in the case of Contextual Response Rephraser) that is stored in the bot utterance event metadata. For rephrased answers, you must specify utter_name to run the assertion.

  These assertions can be specified for user steps only and cannot be used alongside the former E2E test format. You can learn more about this new feature in the documentation sections for grounded and relevant assertion types.

  To enable this feature, please set the environment variable RASA_PRO_BETA_E2E_ASSERTIONS to true.

  export RASA_PRO_BETA_E2E_ASSERTIONS=true

• You can now produce a coverage report of your e2e tests via the following command:

  rasa test e2e <e2e-test-folder> --coverage-report [--coverage-output-path <output-folder>]

  The coverage report contains the number of steps and the number of tested steps per flow. Untested steps are referenced by line numbers.

  Flow Name Coverage  Num Steps  Missing Steps  Line Numbers
     flow_1    0.00%          1              1       [10-10]
     flow_2  100.00%          4              0            []
      Total   80.00%          5              1

  Additionally, we also create a histogram of command coverage showing how many and what commands are produced in your e2e tests.

  To enable this feature, please set the environment variable RASA_PRO_BETA_FINETUNING_RECIPE to true.

  export RASA_PRO_BETA_FINETUNING_RECIPE=true

  More information can be found on the documentation of the feature.

• Create a self-hosted LLM client compatible with OpenAI format. Users can connect to their own self-hosted LLM server that is compatible with OpenAI format.

  Sample basic usage:

      llm:
          provider: self-hosted
          model: <deployment_name>
          api_base: <deployment_url>
          api_type: openai [Optional]

• Add a new CLI command rasa llm finetune prepare-data to create a dataset from e2e tests that can be used to fine-tune a base model for the task of command generation.

  To enable this feature, please set the environment variable RASA_PRO_BETA_FINETUNING_RECIPE to true.

  export RASA_PRO_BETA_FINETUNING_RECIPE=true

• It is now allowed to link to pattern_human_handoff from any pattern and user flow.

• Allow links from all patterns to user flows except for pattern_internal_error.

• • LiteLLM Integration & Reduced LangChain Reliance:
    • Introduced LLMClient and EmbeddingClient protocols for standardized client interfaces.
    • Created lightweight client wrappers for LiteLLM to streamline model instantiation, management, and inference.
    • Updated llm_factory and embedder_factory to utilize these LiteLLM client wrappers.
    • Added dedicated clients for Azure OpenAI and OpenAI to support both LLMs and embedding models.
    • Added a HuggingFace client to compute embeddings using locally stored transformer models via the sentence-transformers package.
  • LangChain Update: Upgraded to the latest version (0.2.x) for improved compatibility and features. To understand the implications on your assistant, please refer to the feature documentation and the migration guide.

• Implement as part of E2E testing a new type of evaluation specifically designed to increase confidence in CALM. This evaluation runs assertions on the assistant's actual events and generative responses. New assertions include the ability to check for the presence of specific events, such as:

  • flow started, flow completed or flow cancelled events
  • whether pattern_clarification was triggered for specific flows
  • whether buttons rendered well as part of the bot uttered event
  • whether slots were set correctly or not
  • whether the bot text response matches a provided regex pattern
  • whether the bot response matches a provided domain response name

  These assertions can be specified for user steps only and cannot be used alongside the former E2E test format. You can learn more about this new feature in the documentation.

  To enable this feature, please set the environment variable RASA_PRO_BETA_E2E_ASSERTIONS to true.

  export RASA_PRO_BETA_E2E_ASSERTIONS=true

• Configure LLM-as-Judge settings in the llm_as_judge section of the conftest.yml file. These settings will be used to evaluate the groundedness and relevance of generated bot responses. The conftest.yml is discoverable as long as it is in the root directory of the assistant project, at the same level as the config.yml file.

  If the conftest.yml file is not present in the root directory, the default LLM judge settings will be used.

• Implement automatic E2E test case conversion from sample conversation data.

  This feature includes:

  • A CLI command to convert sample conversation data (CSV, XLSX) into executable E2E test cases.
  • Conversion of sample data using an LLM to generate YAML formatted test cases.
  • Export of generated test cases into a specified YAML file.

  Usage:

  rasa data convert e2e <path>

  To enable this feature, please set the environment variable RASA_PRO_BETA_E2E_CONVERSION to true.

  export RASA_PRO_BETA_E2E_CONVERSION=true

  For more details, please refer to this documentation page.

Improvements

• Implemented custom action stubbing for E2E test cases. To define custom action stubs, add stub_custom_actions to the test case file.

  Stubs can be defined in two ways:

  • Test file level: Define each action by its name (action_name).
  • Test case level: Define the stub using the test case ID as a prefix (test_case_id::action_name).

  To learn more about this feature, please refer to the documentation.

  To enable this feature, set the environment variable RASA_PRO_BETA_STUB_CUSTOM_ACTION to true:

  export RASA_PRO_BETA_STUB_CUSTOM_ACTION=true

• Add max_messages_in_query parameter to Enterprise Search Policy, it allows controlling the number of past messages that are used in the search query for retrieval

• Configure LLM E2E test converter settings in the llm_e2e_test_conversion section of the conftest.yml file.

  These settings will be used to configure the LLM used to convert sample conversation data into E2E test cases.

  The conftest.yml is discoverable as long as it is in the root directory of the tests output path.

  If the conftest.yml file is not present in the root directory, the default LLM settings will be used.

• Add the datetime of Rasa Pro license expiry to rasa --version command Add /license API endpoint that also returns the same information

• Suppress LiteLLM info and debug log messages in the console.

• Cache llm_factory and embedder_factory methods to avoid client instantiation and validation for every user utterance.

• Added E2E Test Conversion Completed telemetry event with file type and test case count properties.

• Separate writing of failed and passed e2e test results to distinct file paths.

• Implement support for evaluating IntentlessPolicy responses with generative response assertions.

• Use direct custom action execution in tutorial and CALM templates. Skip action server health check in e2e testing if direct custom action execution is configured.

• Modified the type of flows which are included into the import CLI (previously only user flows were enabled, now patterns are included). Use case: This is needed for Studio 1.7, since that release is enabling modification and management of patterns inside Studio, and needs the ability to import patterns from yaml files.

• Improve events and responses sub-schemas used by the stub_custom_actions sub-schema of end-to-end testing. The events sub-schema only allows the usage of events which are supported by the rasa-sdk. These are documented in the action server API documentation.

• Change default model of conversation rephraser to 'gpt-4o-mini'.

• Add file_path to Flow so that we can show the full name, e.g. path/to/flow.py::flow name in the e2e test coverage report.

• Introduced remote storage to upload trained model to persistors(AWS, GCP, Azure)

• Add ability to download training data from remote storage(gcs, aws, azure)

• Allow saving models to and retrieving from sub folders in cloud storage.

• Introduced DirectCustomActionExecutor for executing custom actions directly through the assistant.

  Introduced actions_module variable under action_endpoint in endpoints.yml to explicitly specify the path to custom actions module.

  If actions_module is set, custom actions will be executed directly through the assistant.

• Add validation for the values against which categorical and boolean slots are checked in the if conditional steps. An error will be thrown when a slot is compared to an invalid/non-existent value for boolean and categorical slots.

• Add user query and retrieved document results to the metadata of action_send_text predicted by EnterpriseSearchPolicy. In addition, add domain ground truth responses to the BotUttered event metadata when rephrasing is enabled. These changes were required to allow evaluations of generative responses against the ground truth stored in the metadata of BotUttered events.

Bugfixes

• Fix problem with custom action invocation when model is loaded from remote storage.

• Ensure certificates for openai based clients.

• Mark the first slot event as seen when the user turn in a E2E test case contains multiple slot events for the same slot. This fixes the issue when the assertion_order_enabled is set to true and the user step in a test case contained multiple slot_was_set assertions for the same slot, the last slot event was marked as seen when the first assertion was running. This caused the test to fail for subsequent slot_was_set assertions for the same slot with error Slot <slot_name> was not set.

• Validate the LLM configuration during training for the following components:

  • Contextual Response Rephraser
  • Enterprise Search Policy
  • Intentless Policy
  • LLM Based Command Generator
  • LLM Based Router

  Additionally, update the get_provider_from_config method to retrieve the provider using both the model and model_name configuration parameters.

• Fixes throwing the deprecation warning if the setting for Azure OpenAI Embedding Client was not set through the deprecated environment variable.

• Fix execution of stub custom actions when they contain test case name and the separator in its provided stub name. Test runner will now correctly execute the correct stub implementation for the same custom action dependent on the test name.

• Add validation to conversation rephraser.

• Ensure YAML files with datetime-formatted strings are read as plain strings instead of being converted to datetime objects.

• Deprecate 'request_timeout' for OpenAI and Azure OpenAI clients in favor of 'timeout'

• Forbid stream and n parameters for clients. Having these parameters within llm and embeddings configuration will result in error.

• Raise deprecation warning if api_type is set to huggingface instead of huggingface_local for HuggingFace local embeddings.

• Fix resolving aliases for deprecated keys when instantiating LLM and embedding clients.

• Fix detection of conftest file which contained custom LLM judge configuration.

• Fix issue with Rasa Pro Studio download command exporting default flows which had not been customized by the Studio user. Rasa Pro Studio download command only exports user defined flows, customized patterns and user defined domain locally from the Studio instance.

  Similarly, fix issue with Rasa Pro Studio upload command importing default flows which had not been customized to Studio. Rasa Pro Studio upload command only imports user defined flows, customized patterns and user defined domain to the Studio instance.

• Disable auto-inferring provider from the config. Ensure the provider is explicitly read from the provider key.

• Fix writing e2e test cases to disk. slot_was_set and slot_was_not_set are now written down correctly.

• The rephraser of the rasa llm finetune data-prepare command now compares the original user message and the user message returned in the LLM output case-insensitive.

• [rasa llm finetune prepare-data] Do not rephrase user messages that come from a button payload.

• Separate commands in the expected LLM output by newlines.

• Fix TypeError in PatternClarificationContainsAssertion hash function by converting sets to lists for successful JSON serialization.

• Fix validation in case a link to pattern_human_handoff is used.

• [rasa llm finetune prepare-data] Skip paraphrasing module in case num-rephrases is set to 0.

• Update the handling of incorrect use of slash syntax. Messages with undefined intents do not automatically trigger pattern_cannot_handle; instead, they are sanitized (prepended slash(es) are removed) and passed through the graph.

• Allow suitable patterns to be properly started using nlu triggers

• Fix API connection error for bedrock embedding endpoint.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.9.20] - 2025-04-14

Rasa Pro 3.9.20 (2025-04-14)

Bugfixes

• Updated Inspector dependent packages (cross-spawn, mermaid, dom-purify, vite, braces, ws, axios and rollup) to address security vulnerabilities.
• Modify Enterprise Search Citation Prompt Template to use doc.text
• Security patch for Audiocodes channel connector. Audiocodes channel was only checking for the existence of authentication token. It now does a constant-time string comparison of the authentication token in connection request with that provided in channel configruation.

[3.9.19] - 2025-01-30

Rasa Pro 3.9.19 (2025-01-30)

Improvements

• Remove unnecessary deepcopy to improve performance in undo_fallback_prediction method of FallbackClassifier

Bugfixes

• • Fixed an issue where the pattern_continue_interrupted was not correctly triggered when the flow digressed to a step containing a link.
• Add the flow ID as a prefix to step ID to ensure uniqueness. This resolves a rare bug where steps in a child flow with a structure similar to those in a parent flow (using a "call" step) could result in duplicate step IDs. In this case duplicates previously caused incorrect next step selection.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.9.18] - 2025-01-15

Rasa Pro 3.9.18 (2025-01-15)

Bugfixes

• Fix issue in e2e testing when customising action_session_start would lead to AttributeError, because the output_channel was not set. This is now fixed by setting the output_channel to CollectingOutputChannel().
• Pass flow human-readable name instead of flow id when the cancel pattern stack frame is pushed during flow policy validation checks of collect steps.
• Make pattern_session_start work with rasa inspector to allow the assistant proactively start the conversation with a user.
• Fixes a critical security vulnerability with jsonpickle dependency by upgrading to the patched version.
• Updated minio to address security vulnerability.

[3.9.17] - 2024-12-05

Rasa Pro 3.9.17 (2024-12-05)

Bugfixes

• Implement eq and hash functions for ChangeFlowCommand to fix error=unhashable type: 'ChangeFlowCommand' error in MultiStepCommandGenerator.

[3.9.16] - 2024-11-26

Rasa Pro 3.9.16 (2024-11-26)

Bugfixes

• Replace pickle and joblib with safer alternatives, e.g. json, safetensors, and skops, for serializing components.

  Note: This is a model breaking change. Please retrain your model.

  If you have a custom component that inherits from one of the components listed below and modified the persist or load method, make sure to update your code. Please contact us in case you encounter any problems.

  Affected components:

  • CountVectorFeaturizer
  • LexicalSyntacticFeaturizer
  • LogisticRegressionClassifier
  • SklearnIntentClassifier
  • DIETClassifier
  • CRFEntityExtractor
  • TrackerFeaturizer
  • TEDPolicy
  • UnexpectedIntentTEDPolicy

[3.9.15] - 2024-10-18

Rasa Pro 3.9.15 (2024-10-18)

Improvements

• Change default response of utter_free_chitchat_response from "placeholder_this_utterance_needs_the_rephraser" to "Sorry, I'm not able to answer that right now.".

Bugfixes

• Fix cleanup of SetSlot commands issued by the LLM-based command generator for slots that define a slot mapping other than the from_llm slot mapping. The command processor now correctly removes the SetSlot command in these scenarios and instead adds a CannotHandleCommand.
• Disallow using the command payload syntax to set slots not filled by any of the active or startable flow(s) collect steps.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.9.14] - 2024-10-02

Rasa Pro 3.9.14 (2024-10-02)

No significant changes.

[3.9.13] - 2024-10-01

Rasa Pro 3.9.13 (2024-10-01)

Bugfixes

• Fix inconsistent recording of telemetry events for llm-based command generators.
• Added tracing explicitly to GRPCCustomActionExecutor.run in order to pass the tracing context to the action server.
• Fixes an issue where the CountVectorsFeaturizer and LogisticRegressionClassifier would throw error during inference when no NLU training data is provided.

[3.9.12] - 2024-09-20

Rasa Pro 3.9.12 (2024-09-20)

Deprecations and Removals

• Dropped support for Python 3.8 ahead of Python 3.8 End of Life in October 2024. In Rasa Pro versions 3.10.0, 3.9.11 and 3.8.13, we needed to pin the TensorFlow library version to 2.13.0rc1 in order to remove critical vulnerabilities; this resulted in poor user experience when installing these versions of Rasa Pro with uv pip. Removing support for Python 3.8 will make it possible to upgrade to a stabler version of TensorFlow.

Improvements

• Update Keras and Tensorflow to version 2.14. This will eliminate the need to use the --prerelease allow flag when installing Rasa Pro using uv pip tool.

Bugfixes

• Fix AttributeError with the instrumentation of the run method of the CustomActionExecutor class.
• Fixed UnexpecTEDIntentlessPolicy training errors that resulted from a change to batching behavior. Changed the batching behavior back to the original for all components. Made the changed batching behavior accessible in DietClassifier using drop_small_last_batch: True.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.9.11] - 2024-09-13

Rasa Pro 3.9.11 (2024-09-13)

Bugfixes

• Update Keras to 2.13.1 and Tensorflow to 2.13.0rc0 to fix critical vulnerability (CVE-2024-3660).

[3.9.10] - 2024-09-12

Rasa Pro 3.9.10 (2024-09-12)

Bugfixes

• Fix FileNotFound error when running rasa studio commands and no pre-existing local assistant project exists.
• Fixed telemetry collection for the components Rephraser, LLM Intent Classifier, Intentless Policy and Enterprise Search Policy to ensure that the telemetry data is only collected when it is enabled

[3.9.9] - 2024-08-23

Rasa Pro 3.9.9 (2024-08-23)

Bugfixes

• Updated behaviour of policies in coexistence:

  • CALM policies run in case the routing slot is set to True (routing to CALM).
  • Policies of the nlu-based system run in case the routing slot is set to False (routing to NLU-based system) or None (non-sticky routing).

• Don't create an instance of FlowRetrieval in the command generators in case no flows exists.

• Patterns do not count as active flows in MultiStepLLMCommandGenerator anymore.

• Make sure that all e2e test cases in rasa inspector are valid.

• Downloading of CALM Assistants from Studio improved:

  • Downloading CALM assistants from Studio now includes config and endpoints files
  • Downloading CALM assistants from Studio now doesn't require config.yml and data folder to exist

[3.9.8] - 2024-08-21

Rasa Pro 3.9.8 (2024-08-21)

Bugfixes

• Fix problem with custom action invocation when model is loaded from remote storage.

[3.9.7] - 2024-08-15

Rasa Pro 3.9.7 (2024-08-15)

Bugfixes

• Fix extraction of tracing context from the request headers and injection into the Rasa server tracing context.
• YamlValidationException will correctly return line number of the element where the error occurred when line number of that element is not returned by ruamel.yaml (for elements of primitive types, e.g. str, int, etc.), instead of returning the line number of the parent element.
• Updated setuptools to fix security vulnerability.
• Fix tracing context propagation to work for all external service calls.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.9.6] - 2024-08-07

Rasa Pro 3.9.6 (2024-08-07)

Miscellaneous internal changes

Miscellaneous internal changes.

[3.9.5] - 2024-08-01

Rasa Pro 3.9.5 (2024-08-01)

Improvements

• Enabled generative chitchat in the tutorial template with instructions on how to turn it off added to the documentation.

Bugfixes

• Update the usage of time.process_time_ns with time.perf_counter_ns to fix the inconsistencies between duration metrics and trace spans duration.

[3.9.4] - 2024-07-25

Rasa Pro 3.9.4 (2024-07-25)

Bugfixes

• Fix instrumentation not accounting for kwargs that are passed to NLUCommandAdapter.predict_commands.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.9.3] - 2024-07-18

Rasa Pro 3.9.3 (2024-07-18)

Bugfixes

• Refactor the supported remote storage (AWS, GCS, Azure) verification check before downloading Rasa model by fixing the initial implementation which attempted to create the object storage to check existence.
• Fix TypeError: InformationRetrieval.search() got an unexpected keyword argument when tracing is enabled with EnterpriseSearchPolicy.
• Change warning log level to error log level for Validator methods that verify that forms and actions used in stories and rules are present in the domain.

[3.9.2] - 2024-07-09

Rasa Pro 3.9.2 (2024-07-09)

Bugfixes

• Add key-word arguments in the predict_commands method of LLM-based CommandGenerator class to ensure custom components are not impacted by changes to the signature of the base classes.

[3.9.1] - 2024-07-04

Rasa Pro 3.9.1 (2024-07-04)

Bugfixes

• Modify the validation to throw an error for a missing associated action/utterance in a collect step only if the slot does not have a defined initial value.
• Modify the collect step validation in flow executor to trigger pattern_internal_error for a missing associated action/utterance in a collect step only if the slot does not have a defined initial value.

[3.9.0] - 2024-07-03

Rasa Pro 3.9.0 (2024-07-03)

Features

• Introduce a new response button payload format that runs set slot CALM commands directly by skipping the user message processing pipeline.

• Added support for Information Retrieval custom components. It allows Enterprise Search Policy to be used with arbitrary search systems. Custom Information Retrievals can be implemented as a subclass of rasa.core.information_retrieval.InformationRetrieval

• Enable slot filling in a CALM assistant to be configurable:

  • either use NLU-based predefined slot mappings that instructs NLUCommandAdapter to issue SetSlot commands with values extracted from the user input via an entity extractor or intent classifier
  • or use the new predefined slot mapping from_llm which enables LLM-based command generators to issue SetSlot commands If no slot mapping is defined, the default behavior is to use the from_llm slot mapping.

  In case you had been using custom slot mapping type for slots set with the prediction of the LLM-based command generator, you need to update your assistant configuration to use the new from_llm slot mapping type. Note that even if you have written custom slot validation actions (following the validate_<slot_name> convention) for slots set by the LLM-based command generator, you need to update your assistant configuration to use the new from_llm slot mapping type.

  For slots that are set only via the custom action e.g. slots set by external sources only, you need to add the action name to the slot mapping:

  slots:
    slot_name:
      type: text
      mappings:
        - type: custom
          action: custom_action_name

• Skip SetSlot commands issued by LLM based command generators for slots with NLU-based predefined slot mappings. Instead, the command processor component will issue CannotHandle command to trigger pattern_cannot_handle if no other valid command is found.

• Rasa now supports gRPC protocol for custom actions. This allows users to use gRPC to invoke custom actions. To connect to the action server using gRPC, specify:

  endpoints.yml

  action_endpoint:
      url: "grpc://<rasa-grpc-action-server>:<port>"

  Users can use secure (TLS) and insecure connections to communicate over gRPC. To use TLS specify the following in endpoints.yml:

  endpoints.yml

  action_endpoint:
      url: "grpc://<rasa-grpc-action-server>:<port>"
      cafile: "<ca_file>"

• Add MultiStepLLMCommandGenerator as an alternative LLM based command generator. MultiStepLLMCommandGenerator breaks down the task of dialogue understanding into two steps: handling the flows and filling the slots. The component was designed to enable cheaper and smaller LLMs, such as gpt-3.5-turbo, as viable alternatives to costlier but more powerful models such as gpt-4. To use the MultiStepLLMCommandGenerator add it to your pipeline:

  pipeline:
    ...
    - name: MultiStepLLMCommandGenerator
    ...

Improvements

• Improve diagram display in the inspector by adding an horizontal scroll and an auto scroll to the active step.

• Create a separate default prompt for Enterprise Search with source citation enabled and revert the default Enterprise Search prompt to that of 3.7.x.

• Refactored RemoteAction to utilize a new CustomActionExecutor interface by implementing HTTPCustomActionExecutor to handle HTTP requests for custom actions.

• Implemented an optimization to reduce payload size by ensuring the Assistant sends the domain dictionary to the Action Server only once, which the server then stores. If the Action Server responds with a 449 status code indicating a missing domain context, the Assistant will repeat the API request including the domain dictionary in the payload, ensuring the server properly saves this data.

• Integrate the capability of testing scenarios that reflect actual operational environments where conversations can be influenced by real-time external data. This is done by injecting metadata when running end-to-end tests.

• Introduced LRU caching for reading and parsing YAML files to enhance performance by avoiding multiple reads of the same file. Added READ_YAML_FILE_CACHE_MAXSIZE environment variable with a default value of 256 to configure the cache size.

• Add validations for flow ID to allow only alphanumeric characters, underscores, and hyphens except for the first character.

• The LLMCommandGenerator component has been renamed to SingleStepLLMCommandGenerator. There is no change to the functionality.

  Using the LLMCommandGenerator as the name of the component results in a deprecation warning as it will be permanently renamed to SingleStepLLMCommandGenerator in 4.0.0. Please modify the assistant’s configuration to use the SingleStepLLMCommandGenerator instead of the LLMCommandGenerator to avoid seeing the deprecation warning.

• Make improvements to rasa data validate that check if the usage of slot mappings in a CALM assistant is valid:

  • a slot cannot have both a from_llm mapping and either a nlu-predefined mapping or a custom slot mapping
  • a slot collected in a flow by a custom action has an associated action_ask_ defined in the domain
  • a CALM assistant with slots that have nlu-based predefined mappings include NLUCommandAdapter in the config pipeline
  • a NLU-based assistant cannot have slots that have a from_llm mapping

• Modify post processing of commands - Clarify command with single option is converted into a StartFlow command.

• Improve debug logging for predicate evaluation.

Bugfixes

• Properly handle projects where rasa studio download is run in a project with no NLU data.
• Tracing is supported for actions called over gRPC protocol.
• Fix the hash function of ClarifyCommand to return a hashed list of options.
• Raise an error if action_reset_routing is used without the defined ROUTE_TO_CALM_SLOT / router.
• Add a few bugfixes to the CALM slot mappings feature:
  • Coexistence bot should ignore NoOpCommand when checking if the processed message contains commands.
  • Update condition under which FlowPolicy triggers pattern_internal_error for slots with custom slot mappings.
• Remove invalid warnings during collect step.
• • Fixed issue where messages with invalid intent triggers ('/<intent>') were not handled correctly. Now triggering the pattern_cannot_handle.
  • Introduced a new reason cannot_handle_invalid_intent for use in the pattern_cannot_handle switch mechanism to improve error handling.
• Validates that a collect step in a flow either has an action or an utterance defined in the domain to avoid the bot being silent.
• Slots that are set via response buttons should not trigger pattern_cannot_handle regardless of the slots' mapping type.
• Coerce "None", "null" or "undefined" slot values set via response buttons to be of type NoneType instead of str.
• Avoid raising a UserWarning during validation of response buttons which contain double curly braces.
• Do not run NLUCommandAdapter during message parsing when receiving a /SetSlots button payload. This is because the NLUCommandAdapter run during message parsing (when the graph is skipped) is meant to handle intent button payloads only.
• Exclude slots that are not collected in any flow from being set by the NLUCommandAdapter in a coexistence assistant.
• Default action action_extract_slots should not run custom actions specified in custom slot mappings for slots that are set by custom actions in the flows/CALM system of a coexistence assistant.
• Fix pattern flows being unavailable during input preparation and template rendering in MultiStepLLMCommandGenerator.
• Skip command cleaning when no commands are present in NLUCommandAdapter. Fix get active flows to return the correct active flows, including all the nested parent flows if present.
• If FlowPolicy tries to collect a slot with a custom slot mapping without the action key or action_ask specified in the domain, it will trigger pattern_cancel_flow first, then pattern_internal_error.
• Cancel user flow in progress and invoke pattern_internal_error if the flow reached a collect step which does not have an associated utter_ask response or action_ask action defined in the domain.
• IntentlessPolicy abstains from making a prediction during coexistence when it's the turn of the NLU-based system.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.8.17] - 2024-10-18

Rasa Pro 3.8.17 (2024-10-18)

Improvements

• Change default response of utter_free_chitchat_response from "placeholder_this_utterance_needs_the_rephraser" to "Sorry, I'm not able to answer that right now.".

[3.8.16] - 2024-10-02

Rasa Pro 3.8.16 (2024-10-02)

No significant changes.

[3.8.15] - 2024-10-01

Rasa Pro 3.8.15 (2024-10-01)

Bugfixes

• Fixes an issue where the CountVectorsFeaturizer and LogisticRegressionClassifier would throw error during inference when no NLU training data is provided.

[3.8.14] - 2024-09-20

Rasa Pro 3.8.14 (2024-09-20)

Deprecations and Removals

• Dropped support for Python 3.8 ahead of Python 3.8 End of Life in October 2024. In Rasa Pro versions 3.10.0, 3.9.11 and 3.8.13, we needed to pin the TensorFlow library version to 2.13.0rc1 in order to remove critical vulnerabilities; this resulted in poor user experience when installing these versions of Rasa Pro with uv pip. Removing support for Python 3.8 will make it possible to upgrade to a stabler version of TensorFlow.

Improvements

• Update Keras and Tensorflow to version 2.14. This will eliminate the need to use the --prerelease allow flag when installing Rasa Pro using uv pip tool.

Bugfixes

• Fixed UnexpecTEDIntentlessPolicy training errors that resulted from a change to batching behavior. Changed the batching behavior back to the original for all components. Made the changed batching behavior accessible in DietClassifier using drop_small_last_batch: True.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.8.13] - 2024-09-12

Rasa Pro 3.8.13 (2024-09-12)

Bugfixes

• Fixed telemetry collection for the components Rephraser, LLM Intent Classifier, Intentless Policy and Enterprise Search Policy to ensure that the telemetry data is only collected when it is enabled
• Update Keras to 2.13.1 and Tensorflow to 2.13.0rc0 to fix critical vulnerability (CVE-2024-3660).

[3.8.12] - 2024-08-12

Rasa Pro 3.8.12 (2024-08-12)

Bugfixes

• Fix TypeError: InformationRetrieval.search() got an unexpected keyword argument when tracing is enabled with EnterpriseSearchPolicy.
• Fix extraction of tracing context from the request headers and injection into the Rasa server tracing context.
• Update the usage of time.process_time_ns with time.perf_counter_ns to fix the inconsistencies between duration metrics and trace spans duration.
• YamlValidationException will correctly return line number of the element where the error occurred when line number of that element is not returned by ruamel.yaml (for elements of primitive types, e.g. str, int, etc.), instead of returning the line number of the parent element.
• Updated setuptools to fix security vulnerability.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.8.11] - 2024-07-04

Rasa Pro 3.8.11 (2024-07-04)

Improvements

• Improve debug logging for predicate evaluation.

Bugfixes

• Raise an error if action_reset_routing is used without the defined ROUTE_TO_CALM_SLOT / router.
• Remove invalid warnings during collect step.
• • Fixed issue where messages with invalid intent triggers ("/intent_name") were not handled correctly. Now triggering the pattern_cannot_handle.
  • Introduced a new reason cannot_handle_invalid_intent for use in the pattern_cannot_handle switch mechanism to improve error handling.
• Validates that a collect step in a flow either has an action or an utterance defined in the domain to avoid the bot being silent.
• Skip command cleaning when no commands are present in NLUCommandAdapter. Fix get active flows to return the correct active flows, including all the nested parent flows if present.
• Update the handling of incorrect use of slash syntax. Messages with undefined intents do not automatically trigger pattern_cannot_handle; instead, they are sanitized (prepended slash(es) are removed) and passed through the graph.
• Modify the validation to throw an error for a missing associated action/utterance in a collect step only if the slot does not have a defined initial value.

[3.8.10] - 2024-06-19

Rasa Pro 3.8.10 (2024-06-19)

Improvements

• Added NLG validation to the rasa model training process.

Bugfixes

• Fixes Clarify command being dropped by command processor due to presence of coexistence slot - route_session_to_calm
• Fix validation for LLMBasedRouter to check only for calm_entry.sticky

[3.8.9] - 2024-06-14

Rasa Pro 3.8.9 (2024-06-14)

Improvements

• Add validations for flow ID to allow only alphanumeric characters, underscores, and hyphens except for the first character.

[3.8.8] - 2024-06-07

Rasa Pro 3.8.8 (2024-06-07)

Bugfixes

• Add wrappers around openai clients that can set the self-signed certs via REQUESTS_CA_BUNDLE env variable.

[3.8.7] - 2024-05-29

Rasa Pro 3.8.7 (2024-05-29)

Bugfixes

• Add support for domain entities in CALM import
• Download NLU data when running rasa studio download for a modern assistant with NLU triggers. Previously, this data was not downloaded, leading to a partial assistant.

[3.8.6] - 2024-05-27

Rasa Pro 3.8.6 (2024-05-27)

Improvements

• Adds tracker_state attribute to OutputChannel. It simplifies the access of tracker state for custom channel connector with CollectingOutputChannel.tracker_state.

Bugfixes

• If a button in a response does not have a payload, socketio channel will use the title as payload by default rather than throwing an exception.

[3.8.5] - 2024-05-03

Rasa Pro 3.8.5 (2024-05-03)

Bugfixes

• Trigger pattern_internal_error if collection does not exist in a Qdrant vector store.

[3.8.4] - 2024-04-30

Rasa Pro 3.8.4 (2024-04-30)

Improvements

• Added support for NLU Triggers by supporting uploading the NLU files for CALM Assistants

[3.8.3] - 2024-04-26

Rasa Pro 3.8.3 (2024-04-26)

Improvements

• • Throw validation error and exit when duplicate responses are found across domains. This is a breaking change, as it will cause training to fail if duplicate responses are found. If you have duplicate responses in your training data, you will need to remove them before training.
  • Update domain importing to ignore the warnings about duplicates when merging with the default flow domain

Bugfixes

• Use AzureChatOpenAI class instead of AzureOpenAI class to instantiate openai models deployed in Azure. This fixes the usage of gpt-3.5-turbo model in Azure.
• Fixes validation to catch empty placeholders in response that dumps entire context.
• Fix security vulnerabilities by updating poetry environment: fonttools, CVE-2023-45139, from 4.40.0 to 4.43.0 aiohttp, CVE-2024-27306, from 3.9.3 to 3.9.4 dnspython, CVE-2023-29483, from 2.3.0 to 2.6.1 pymongo, CVE-2024-21506, from 4.3.3 to 4.6.3
• Numbers that are part of the body of the LLM answer in EnterpriseSearch should not be matched as citation references in the postprocessing method.
• Errors from the Flow Retrieval API are now both logged and thrown. When such errors occur, an ErrorCommand is emitted by the Command Generator.

[3.8.2] - 2024-04-25

Rasa Pro 3.8.2 (2024-04-25)

Bugfixes

• Add the currently active flow as well as the called flow (if present) to the list of available flows for the LLMCommandGenerator.
• Fix custom prompt not read from the model resource path for LLMCommandGenerator.

[3.8.1] - 2024-04-17

Rasa Pro 3.8.1 (2024-04-17)

Improvements

• Adjusted chat widget behavior to remain open when clicking outside the chat box area.
• Improve debug logs to include information about evaluation of if-else conditions in flows at runtime.
• Remove the ContextualResponseRephraser from the tutorial template to keep it simple as it is not needed anymore.
• Update poetry package manager version to 1.8.2. Check the migration guide for instructions on how to update your environment.

Bugfixes

• Introduced support for numbered Markdown lists.
• Added support for uploading assistants with default domain directory.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.8.0] - 2024-04-03

Rasa Pro 3.8.0 (2024-04-03)

Features

• Introduces semantic retrieval of flows at runtime to reduce the size of the prompt sent to the LLM by utilizing similarity between vector embeddings. It enables the assistant to scale to a large number of flows.

  Flow retrieval is enabled by default. To configure it, you can modify the settings under the flow_retrieval property of LLMCommandGenerator component. For detailed configuration options, refer to our documentation.

  Introduces always_include_in_prompt field to the flow definition. If field is set to true and the flow guard defined in the if field evaluates to true, the flow will be included in the prompt.

• Introduction of coexistence between CALM and NLU-based assistants. Coexistence allows you to use policies from both CALM and NLU-based assistants in a single assistant. This allows migrating from NLU-based paradigm to CALM in an iterative fashion.

• Introduction of call step. You can use a call step to embed another flow. When the execution reaches a call step, Rasa starts the called flow. Once the called flow is complete, the execution continues with the calling flow.

Improvements

• Instrument the command_processor module, in particular the following functions:

  • execute_commands
  • clean_up_commands
  • validate_state_of_commands
  • remove_duplicated_set_slots

• Improve the instrumentation of LLMCommandGenerator:

  • extract more LLM configuration parameters, e.g. type, temperature, request-timeout, engine and deployment (the latter 2 being only for the Azure OpenAI service).
  • instrument the private method _check_commands_against_startable_flows to track the commands with which the LLM responded, as well as the startable flow ids.

• Instrument flow_executor.py module, in particular these functions:

  • advance_flows(): extract available_actions tracing tag
  • advance_flows_until_next_action(): extract action name and score, metadata and prediction events as tracing tags from the returned prediction value
  • run_step(): extract step custom id, description and current flow id.

• Instrument Policy._prediction() method for each of the policy subclasses.

• Instrument IntentlessPolicy methods such as:

  • find_closest_response: extract the response and score from the returned tuple;
  • select_response_examples: extract the ai_response_examples from returned value;
  • select_few_shot_conversations: extract the conversation_samples from returned value;
  • extract_ai_responses: extract the ai_responses from returned value;
  • generate_answer: extract the llm_response from returned value.
• 1. Instrument InformationRetrieval.search method for supported vector stores: extract query and document metadata tracing attributes.
  2. Instrument EnterpriseSearchPolicy._generate_llm_answer method: extract LLM config tracing attributes.
  3. Extract dialogue stack current context in the following functions:
  • rasa.dialogue_understanding.processor.command_processor.clean_up_commands
  • rasa.core.policies.flows.flow_executor.advance_flows
  • rasa.core.policies.flows.flow_executor.run_step
• 1. Instrument NLUCommandAdapter.predict_commands method and extract the commands from the returned value, as well as the user message intent.
  2. Improve LLM config tracing attribute extraction for ContextualResponseRephraser.

• Add new config boolean property trace_prompt_tokens that would enable the tracing of the length of the prompt tokens for the following components:

  • LLMCommandGenerator
  • EnterpriseSearchPolicy
  • IntentlessPolicy
  • ContextualResponseRephraser

• Enable execution of single E2E tests by including the test case name in the path to test cases, like so: path/to/test_cases.yml::test_case_name or path/to/folder_containing_test_cases::test_case_name.

• Implement MetricInstrumentProvider interface whose role is to:

  • register instruments during metrics configuration
  • retrieve the appropriate instrument to record measurements in the relevant instrumentation code section

• Enabled the setting of a minimum similarity score threshold for retrieved documents in Enterprise Search's vector_store with the addition of the threshold property. If no documents are retrieved, it triggers Pattern Cannot Handle. This feature is supported in Milvus and Qdrant vector stores.

• Record measurements for the following metrics in the instrumentation code:

  • CPU usage of the LLMCommandGenerator
  • memory usage of LLMCommandGenerator
  • prompt token usage of LLMCommandGenerator
  • method call duration for LLM specific calls (in LLMCommandGenerator, EnterpriseSearchPolicy, IntentlessPolicy, ContextualResponseRephraser)
  • rasa client request duration
  • rasa client request body size

  Instrument EndpointConfig.request() method call in order to measure the client request metrics.

• Improvements around default behaviour of ChitChatAnswerCommand():

  • The command processor will issue CannotHandleCommand() instead of the ChitChatCommand() when pattern_chitchat uses an action step action_trigger_chitchat without the IntentlessPolicy being configured. During training a warning is raised.
  • Changed the default pattern_chitchat to:

  pattern_chitchat:
    description: handle interactions with the user that are not task-oriented
    name: pattern chitchat
    steps:
      - action: action_trigger_chitchat
  • Default rasa init template for CALM comes with IntentlessPolicy added to pipeline.

• Add support for OTLP Collector as metrics receiver which can forward metrics to the chosen metrics backend, e.g. Prometheus.

• Enable document source citation for Enterprise Search knowledge answers by setting the boolean citation_enabled: true property in the config.yml file:

  policies:
    - name: EnterpriseSearchPolicy
      citation_enabled: true

• Add telemetry events for flow retrieval and call step

• Tighten python dependency constraints in pyproject.toml, hence reducing the installation time to around 20 minutes with pip (and no caching enabled).

• Improved tracing clarity of the Contextual Response Rephraser by adding the _create_history method span, including its LLM configuration attributes.

• Users now have enhanced control over the debugging process of LLM-driven components. This update introduces a fine-grained, customizable logging that can be controlled through specific environment variables.

  For example, set the LOG_LEVEL_LLM environment variable to enable detailed logging at the desired level for all the LLM components or specify the component you are debugging:

  Example configuration

  export LOG_LEVEL_LLM=DEBUG
  export LOG_LEVEL_LLM_COMMAND_GENERATOR=INFO
  export LOG_LEVEL_LLM_ENTERPRISE_SEARCH=INFO
  export LOG_LEVEL_LLM_INTENTLESS_POLICY=DEBUG
  export LOG_LEVEL_LLM_REPHRASER=DEBUG

• If the user wants to chat with the assistant at the end of rasa init, we are now calling rasa inspect instead of rasa shell.

• A slot can now be collected via an action action_ask_<slot-name> instead of the utterance utter_ask_<slot-name> in a collect step. You can either define an utterance or an action for the collect step in your flow. Make sure to add your custom action action_ask_<slot-name> to the domain file.

• Validate the configuration of the coexistence router before the actual training starts.

• Improved error handling in Enterprise Search Policy, changed the prompt to improve formatting of documents and ensured empty slots are not added to the prompt.

• Implement asynchronous graph execution. CALM assistants rely on a lot of I/O calls (e.g. to a LLM service), which impaired performances. With this change, we've improved the response time performance by 10x. All policies and components now support async calling.

• Merge rasa and rasa-plus packages into one. As a result, we renamed the Python package to rasa-pro and the Docker image to rasa-pro. Please head over to the migration guide here for installation, and here for the necessary configuration updates.

Bugfixes

• Updated pillow and jinja2 packages to address security vulnerabilities.

• Fix OpenTelemetry Invalid type NoneType for attribute value warning.

• Add support for metadata_payload_key for Qdrant Vector Store with an error message if content_payload_key or metadata_payload_key are incorrect

• Changed the ordering of returned events to order by ID (previously timestamp) in SQL Tracker Store

• Improved the end-to-end test comparison mechanism to accurately handle and strip trailing newline characters from expected bot responses, preventing false negatives due to formatting mismatches.

• Fixed a bug that caused inaccurate search results in Enterprise Search when a bot message appeared before the last user message.

• Fixes flow guards pypredicate evaluatation bug: pypredicate was evaluated with Slot instances instead of slot values

• Post-process source citations in Enterprise Search Policy responses so that they are enumerated in the correct order.

• Resolves issue causing the FlowRetrieval.populate to always use default embeddings.

• Fix the bug with the validation of routing setup crashing when the pipeline is not specified (null)

• Remove conversation turns prior to a restart when creating a conversation transcript for an LLM call.

  This helps in cases where the prior conversation is not relevant for the current session. Information which should be carried to the next session should explicitly be stored in slots.

• Add tracker back to the LLMCommandGenerator.parse_command to ensure compatibility with custom command generator built with 3.7.

• Move coexistence routing setup validation from rasa.validator.Validator to rasa.engine.validation. This gave access to graph schema which allowed for validation checks of subclassed routers.

• Fixes a bug in determining the name of the model based on provided parameters.

• LogisticRegressionClassifier checks if training examples are present during training and logs a warning in case no training examples are provided.

• Fixes the bug that resulted in an infinite loop on a collect step in a flow with a flow guard set to if: False.

• Fix training the enterprise search policy multiple times with a different source folder name than the default name "docs".

• Log message llm_command_generator.predict_commands.finished is set to debug log by default. To enable logging of the LLMCommandGenerator set LOG_LEVEL_LLM_COMMAND_GENERATOR to INFO.

• Improvements and fixes to cleaning up commands:

  • Clean up predicted StartFlow commands from the LLMCommandGenerator if the flow, that should be started, is already active.
  • Clean up predicted SetSlot commands from the LLMCommandGenerator if the value of the slot is already set on the tracker.
  • Use string comparison for slot values to make sure to capture cases when the LLMCommandGenerator predicted a string value but the value set on the tracker is, for example, an integer value.

• Remove context from list of restricted slots

• Improved handling of categorical slots with text values when using CALM.

  Slot values extracted by the command generator (LLM) will be stored in the same casing as the casing used to define the categorical slot values in the domain. E.g. A categorical slot defined to store the values ["A", "B"] will store "A" if the LLM predicts the slot to be filled with "a". Previously, this would have stored "a".

Miscellaneous internal changes

Miscellaneous internal changes.

[3.7.9] - 2024-03-26

Rasa Pro 3.7.9 (2024-03-26)

Improvements

• Add validations for flow ID to allow only alphanumeric characters, underscores, and hyphens except for the first character.

Bugfixes

• Changed the ordering of returned events to order by ID (previously timestamp) in SQL Tracker Store

• Fixes flow guards pypredicate evaluatation bug: pypredicate was evaluated with Slot instances instead of slot values

• Improved handling of categorical slots with text values when using CALM.

  Slot values extracted by the command generator (LLM) will be stored in the same casing as the casing used to define the categorical slot values in the domain. E.g. A categorical slot defined to store the values ["A", "B"] will store "A" if the LLM predicts the slot to be filled with "a". Previously, this would have stored "a".

• Log message llm_command_generator.predict_commands.finished is set to debug log by default. To enable logging of the LLMCommandGenerator set LOG_LEVEL_LLM_COMMAND_GENERATOR to INFO.

• Improvements and fixes to cleaning up commands:

  • Clean up predicted StartFlow commands from the LLMCommandGenerator if the flow, that should be started, is already active.
  • Clean up predicted SetSlot commands from the LLMCommandGenerator if the value of the slot is already set on the tracker.
  • Use string comparison for slot values to make sure to capture cases when the LLMCommandGenerator predicted a string value but the value set on the tracker is, for example, an integer value.

[3.7.8] - 2024-02-28

Rasa Pro 3.7.8 (2024-02-28)

Improvements

• Improved UX around ClarifyCommand by checking options for existence and ordering them. Also, now dropping Clarify commands if there are any other commands to prevent two questions or statements to be uttered at the same time.
• LLMCommandGenerator returns CannotHandle() command when is encountered with scenarios where it is unable to predict a valid command.

Bugfixes

• Replace categorical slot values in a predicate with lower case replacements. This fixes the case sensitive slot comparisons in flow guards, branches in flows and slot rejections.
• Modify flows YAML schema to make next step mandatory to noop step.
• Flush messages when Kafka producer is closed. This is to ensure that all messages in the producer's internal queue are sent to the broker. Ensure to import all pattern stack frame subclasses of DialogueStackFrame when retrieving tracker from the tracker store, a required step during rasa export.
• Add support for metadata_payload_key for Qdrant Vector Store with an error message if content_payload_key or metadata_payload_key are incorrect

[3.7.7] - 2024-02-06

Rasa Pro 3.7.7 (2024-02-06)

Bugfixes

• Updated pillow and jinja2 packages to address security vulnerabilities.

[3.7.6] - 2024-02-01

Rasa Pro 3.7.6 (2024-02-01)

Bugfixes

• Fix reported issue, e.g. https://github.com/RasaHQ/rasa/issues/5461 in Rasa Pro: Do not unpack json payload if data key is not present in the response custom output payloads when using socketio channel. This allows assistants which use custom output payloads to work with the Rasa Inspector debugging tool.
• Make flow description a required property in the flow json schema.
• Fix training the enterprise search policy multiple times with a different source folder name than the default name "docs".

Miscellaneous internal changes

Miscellaneous internal changes.

[3.7.5] - 2024-01-24

Rasa Pro 3.7.5 (2024-01-24)

Improvements

• Add new embedding types: huggingface and huggingface_bge. These new types import the HuggingFaceEmbeddings and HuggingFaceBgeEmbeddings embedding classes from Langchain.

Bugfixes

• Fixes a bug that caused the full_retrieval_intent_name key to be missing in the published event. Rasa Analytics makes use of this key to get the Retrieval Intent Name
• Pin grpcio indirect dependency to 1.56.2 to address CVE-2023-33953 Pin aiohttp to version 3.9.0 to address CVE-2023-49081
• Fixes the bug that resulted in an infinite loop on a collect step in a flow with a flow guard set to if: False.
• Changed the parameters request timeout to 10 seconds and maximum number of retries to 1 for the default LLM used by Enterprise Search Policy. Any error during vector search or LLM API calls should now trigger the pattern pattern_internal_error. Updated the default enterprise search policy prompt to respond more succinctly to queries.

[3.7.4] - 2024-01-03

Rasa Pro 3.7.4 (2024-01-03)

Improvements

• Add embeddings type azure to simplify azure configurations, particularly when using Enterprise Search Policy

Bugfixes

• Add a validation in rasa data validate to check the LinkFlowStep refers to a valid flow ID

[3.7.3] - 2023-12-21

Rasa Pro 3.7.3 (2023-12-21)

Improvements

• Persist prompt as part of the model and reread prompt from the model storage instead of original file path during loading. Impacts LLMCommandGenerator.
• Replaced soon to be depracted text-davinci-003 model with gpt-3.5-turbo. Affects components - LLM Intent Classifier and Contextual Response Rephraser.

Bugfixes

• Fix stale cache of local knowledge base used by EnterpriseSearchPolicy by implementing the fingerprint_addon class method.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.7.2] - 2023-12-07

Rasa Pro 3.7.2 (2023-12-07)

Bugfixes

• Fix propagation of context across rasa spans when running rasa run --enable-api in the case when no additional tracing context is passed to rasa.
• Fixed a bug in policy invocation that made Enterprise Search Policy and action_trigger_search behaved strangely when used with rules and stories
• Updated aiohttp, cryptography and langchain to address security vulnerabilities.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.7.1] - 2023-12-01

Rasa Pro 3.7.1 (2023-12-01)

Improvements

• Improved error handling in Enterprise Search Policy, changed the prompt to improve formatting of documents and ensured empty slots are not added to the prompt

[3.7.0] - 2023-11-22

Rasa Pro 3.7.0 (2023-11-22)

Features

• Added Enterprise Search Policy that uses an LLM with conversation context and relevant knowledge base documents to generate rephrased responses. The LLM is prompted to answer the user questions given the chat transcript, documents retrived from a document search and the slot values so far. This policy supports an in-memory Faiss vector store and connecting to instances of Milvus or Qdrant vector store.

Improvements

• Skip executing the pipeline when the user message is of the form /intent or /intent + entities.

• Remove tensorflow-addons from dependencies as it is now deprecated.

• Add building multi-platform Docker image (amd64/arm64)

• Switch struct log to FilteringBoundLogger in order to retain log level set in the config.

• Added metadata as an additional argument as an additional parameter to an Actions run method.

  Added an additional default action called action_send_text which allows a policy to respond with a text. The text is passed to the action using the metadata, e.g. metadata={"message": {"text": "Hello"}}.

  Added LLM utility functions.

• Passed request headers from REST channel.

• Added additional method fingerprint_addon to the GraphComponent interface to allow inclusion of external data into the fingerprint calculation of a component

• Added Schema file and schema validation for flows.

• Added environment variables to configure JWT and auth token. For JWT the following environment variables are available:

  • JWT_SECRET
  • JWT_METHOD
  • JWT_PRIVATE_KEY

  For auth token the following environment variable is available:

  • AUTH_TOKEN

• Add skip question command

• Update the CALM starter template by:

  • adding the following flows from the financial chatbot:
    • add_contact.yml
    • remove_contact.yml
    • list_contacts.yml
  • using multiple modules (in the form of yml files) to segregate the flows (a good model to be followed)
  • adding e2e tests:
    • happy paths
    • cancelations
    • corrections
• • Enhanced the Rasa error pattern for accommodating various error types.
  • Upgraded the LLMCommandGenerator for processing the new 'user_input' configuration section. This update includes handling of messages that surpass the defined character limit.

  Configuration Update:

  The LLMCommandGenerator now supports a user-defined character limit via the 'user_input' configuration:

    - name: LLMCommandGenerator
      llm:
        ...
      user_input:
        max_characters: 500

  Default Behavior:

  In the absence of a specified limit, it defaults to a 420-character cap. To bypass the limit entirely, set the 'max_characters' value to -1.

• • Bot now returns a default message in response to an empty user message. This improves user experience by providing feedback even when no input is detected.
  • LLMCommandGenerator behavior updated. It now returns an ErrorCommand for empty user messages.
  • Updated default error pattern and added the default utterance in default_flows_for_patterns.yml

• Add support for Vault namespaces. To use namespace set either:

  • VAULT_NAMESPACE environment variable
  • namespace property in secrets_manager section at endpoints.yaml

• Added Rasa Labs LLM components. Added components are:

  • LLMIntentClassifier
  • IntentlessPolicy
  • ContextualResponseRephraser

• Made it possible for the Rasa REST channel to accept OpenTelemetry tracing context.

• Improved the naming of trace spans and added more trace tags.

• Add slot_was_not_set to E2E testing for asserting that a slot was not set and that a slot was not set with a specific value.

• Introduced the rasa studio download command, enabling data retrieval from the studio. Implemented the option to refresh the Keycloak token. Expanded the functionality of RasaPrimitiveStorageMapper with the addition of flows. Added flows support to rasa studio train.

• Instrument LLMCommandGenerator._generate_action_list_using_llm and Command.run_command_on_tracker methods.

• Added the default values for the number of tokens generated by the LLM (max_tokens)

• Make the instrumentation of Command.run_command_on_tracker method applicable to all subclasses of the Command class`

• Instrument ContextualResponseRephraser._generate_llm_response and ContextualResponseRephraser.generate methods.

• Extract commands as tracing attributes from message input when previous node was the LLMCommandGenerator.

• Rename rasa chat command to rasa inspect and rename channel name to inspector.

• Extract events and optional_events when GraphNode is FlowPolicy.

Bugfixes

• uvloop is disabled by default on apple silicon machines

• Add rasa_events to the list of anonymizable structlog keys and rename structlog keys.

• Introduce a validation step in rasa data validate and rasa train commands to identify non-existent paths and empty domains.

• Rich responses containing buttons with parentheses characters are now correctly parsed. Previously any characters found between the first identified pair of () in response button took precedence.

• Resolve dependency incompatibility: Pin version of dnspython to ==2.3.0.

• Fixed KeyError which resulted when domain_responses doesn't exist as a keyword argument while using a custom action dispatcher with nlg server.

• Fixed incompatibility with latest python-socketio release.

  The python-socketio released a backwards incompatible change on their minor release. This fix addresses this and makes the code compatible with prior and the new python-socketio version.

  https://github.com/miguelgrinberg/python-socketio/blob/main/CHANGES.md

• Fixed the 404 Not Found Github actions error while removing packages.

• Corrected E2E diff behavior to prevent it from going out of sync when more than one turn difference exists between actual and expected events. Fixed E2E tests from propagating errors when events and test steps did not have the same length. Fixed the issue where E2E tests couldn't locate slot events that were not arranged chronologically. Resolved the problem where E2E tests were incorrectly diffing user utter events when they were not in the correct order.

• Fixed E2E runner wrongly selecting the first available bot utterance when generating the test fail diff.

• Updated werkzeug and urllib3 to address security vulnerabilities.

• Fix cases when E2E test runner crashes when there is no response from the bot.

Improved Documentation

• Update wording in Rasa Pro installation page.
• Updated docs on sending Conversation Events to Multiple DBs.
• Corrected action server api sample in docs.
• Document support for Vault namespaces.
• Updated tracing documentation to include tracing in the action server and the REST Channel.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.6.13] - 2023-10-23

Rasa Pro 3.6.13 (2023-10-23)

Bugfixes

• Fix wrong conflicts that occur when rasa validate stories is run with slots that have active_loop set to null in mapping conditions.

[3.6.12] - 2023-10-10

Rasa Pro 3.6.12 (2023-10-10)

Improvements

• Added username to the connection parameters for ConcurrentRedisLockStore.

Bugfixes

• Refresh headers used in requests (e.g. action server requests) made by EndpointConfig using its headers attribute.
• Upgrade pillow to 10.0.1 to address security vulnerability CVE-2023-4863 found in 10.0.0 version.
• Fix setuptools security vulnerability CVE-2022-40897 in Docker build by updating setuptools in poetry's environment.

[3.6.11] - 2023-10-05

Rasa Pro 3.6.11 (2023-10-05)

Bugfixes

• Intent names will not be falsely abbreviated in interactive training (fixes OSS-413).

  This will also fix a bug where forced user utterances (using the regex matcher) will be reverted even though they are present in the domain.

• Cache EndpointConfig session object using cached_property decorator instead of recreating this object on every request. Initialize these connection pools for action server and model server endpoints as part of the Sanic after_server_start listener. Also close connection pools during Sanic after_server_stop listener.

[3.6.10] - 2023-09-26

Rasa Pro 3.6.10 (2023-09-26)

Improvements

• Improved handling of last batch during DIET and TED training. The last batch is discarded if it contains less than half a batch size of data.
• Added username to the connection parameters for RedisLockStore and RedisTrackerStore
• Telemetry data is only send for licensed users.

Improved Documentation

• Remove the Playground from docs.

[3.6.9] - 2023-09-15

Rasa Pro 3.6.9 (2023-09-15)

Improvements

• Added additional method fingerprint_addon to the GraphComponent interface to allow inclusion of external data into the fingerprint calculation of a component

Bugfixes

• Fixed KeyError which resulted when domain_responses doesn't exist as a keyword argument while using a custom action dispatcher with nlg server.

[3.6.8] - 2023-08-30

Rasa Pro 3.6.8 (2023-08-30)

Bugfixes

• Fix E2E testing diff algorithm to support the following use cases:
  • asserting a slot was not set under a slot_was_set block
  • asserting multiple slot names and/or values under a slot_was_set block Additionally, the diff algorithm has been improved to show a higher fidelity result.

[3.6.7] - 2023-08-29

Rasa Pro 3.6.7 (2023-08-29)

Bugfixes

• Updated certifi, cryptography, and scipy packages to address security vulnerabilities.
• Updated setuptools and wheel to address security vulnerabilities.

[3.6.6] - 2023-08-23

Rasa Pro 3.6.6 (2023-08-23)

Bugfixes

• Updated setuptools and wheel to address security vulnerabilities.

[3.6.5] - 2023-08-17

Rasa Pro 3.6.5 (2023-08-17)

Improvements

• Use the same session across requests in RasaNLUHttpInterpreter

Bugfixes

• Resolve dependency incompatibility: Pin version of dnspython to ==2.3.0.
• Fix the issue in rasa test e2e where test diff inaccurately displayed actual event transcripts, leading to the duplication of BotUtter`` or UserUtter events. This occurred specifically when `SetSlot events took place that were not explicitly defined in the Test Cases.

Improved Documentation

• Updated PII docs with new section on how to use Rasa X/Enterprise with PII management solution, and a new note on debug logs being displayed after the bot message with rasa shell.

[3.6.4] - 2023-07-21

Rasa Pro 3.6.4 (2023-07-21)

Bugfixes

• Extract conditional response variation and channel variation filtering logic into a separate component. Enable usage of this component in the NaturalLanguageGenerator subclasses (e.g. CallbackNaturalLanguageGenerator, TemplatedNaturalLanguageGenerator). Amend nlg_request_format to include a single response ID string field, instead of a list of IDs.
• Added details to the logs of successful and failed cases of running the markers upload command.

Improved Documentation

• Updated commands with square brackets e.g (pip install rasa[spacy]) to use quotes (pip install 'rasa[spacy]') for compatibility with zsh in docs.

[3.6.3] - 2023-07-20

Rasa Pro 3.6.3 (2023-07-20)

Improvements

• Added a human readable component to structlog using the event_info key and made it the default rendered key if present.

Bugfixes

• Fix the issue with the most recent model not being selected if the owner or permissions where modified on the model file.
• Fixed BlockingIOError which occured as a result of too large data passed to strulogs.
• Fixed the error handling mechanism in rasa test e2e to quickly detect and communicate errors when the action server, defined in endpoints.yaml, is not available.
• Allow hyphens - to be present in e2e test slot names.
• Resolved issues in rasa test e2e where errors occurred when the bot concluded the conversation with SetSlot events while there were remaining steps in the test case. Corrected the misleading error message '- No slot set' to '- Slot types do not match' in rasa test e2e when a type mismatch occurred during testing.

Improved Documentation

• Update action server documentation with new capability to extend Sanic features by using plugins. Update rasa-sdk dependency to version 3.6.1.
• Updated commands with square brackets e.g (pip install rasa[spacy]) to use quotes (pip install 'rasa[spacy]') for compatibility with zsh in docs.

[3.6.2] - 2023-07-06

Rasa Pro 3.6.2 (2023-07-06)

Improvements

• Add building Docker container for arm64 (e.g. to allow running Rasa inside docker on M1/M2).

  Bumped the version of OpenTelemetry to meet the requirement of protobuf 4.x.

Bugfixes

• Resolves the issue of importing TensorFlow on Docker for ARM64 architecture.

[3.6.1] - 2023-07-03

Rasa Pro 3.6.1 (2023-07-03)

Improvements

• Add building multi-platform Docker image (amd64/arm64)
• Switch struct log to FilteringBoundLogger in order to retain log level set in the config.
• Add new anonymizable structlog keys.

Bugfixes

• Add rasa_events to the list of anonymizable structlog keys and rename structlog keys.
• Introduce a validation step in rasa data validate and rasa train commands to identify non-existent paths and empty domains.
• Rich responses containing buttons with parentheses characters are now correctly parsed. Previously any characters found between the first identified pair of () in response button took precedence.
• Add PII bugfixes (e.g. handling None values and casting data types to string before being passed to the anonymizer) after testing manually with Audiocodes channel.

Improved Documentation

• Update wording in Rasa Pro installation page.
• Document new PII Management section.
• Added Documentation for Realtime Markers Section.
• Add "Rasa Pro Change Log" to documentation.
• Document new Load Testing Guidelines section.
• Changes the formatting of realtime markers documentation page

[3.6.0] - 2023-06-14

Rasa Pro 3.6.0 (2023-06-14)

Deprecations and Removals

• Removed Python 3.7 support as it reaches its end of life in June 2023

Features

• Implemented PII (Personally Idenfiable Information) management using Microsoft Presidio as the entity analyzer and anonymization engine. The feature covers the following:

  • anonymization of Rasa events (UserUttered, BotUttered, SlotSet, EntitiesAdded) before they are streamed to Kafka event broker anonymization topics specified in endpoints.yml.
  • anonymization of Rasa logs that expose PII data

  The main components of the feature are:

  • anonymization rules that define in endpoints.yml the PII entities to be anonymized and the anonymization method to be used
  • anonymization executor that executes the anonymization rules on a given text
  • anonymization orchestrator that orchestrates the execution of the anonymization rules and publishes the anonymized event to the matched Kafka topic.
  • anonymization pipeline that contains a list of orchestrators and is registered to a singleton provider component, which gets invoked in hook calls in Rasa Pro when the pipeline must be retrieved for anonymizing events and logs.

  Please read through the PII Management section in the official documentation to learn how to get started.

• Implemented support for real time evaluation of Markers with the Analytics Data Pipeline, enabling you to gain valuable insights and enhance the performance of your Rasa Assistant.

  For this feature, we've added support for rasa markers upload command. Running this command validates the marker configuration file against the domain file and uploads the configuration to Analytics Data Pipeline.

Improvements

• Add optional property ids to the nlg server request body. IDs will be transmitted to the NLG server and can be used to identify the response variation that should be used.

• Add building Docker container for arm64 (e.g. to allow running Rasa inside docker on M1/M2).

• Add support for Location data from Whatsapp on Twilio Channel

• Add validation to rasa train to align validation expectations with rasa data validate. Add --skip-validation flag to disable validation and --fail-on-validation-warnings, --validation-max-history to rasa train to have the same options as rasa data validate.

• Updated tensorflow to version 2.11.1 for all platforms except Apple Silicon which stays on 2.11.0 as 2.11.1 is not available yet

• Slot mapping conditions accept active_loop specified as null in those cases when slots with this mapping condition should be filled only outside form contexts.

• Add an optional description key to the Markers Configuration format. This can be used to add documentation and context about marker's usage. For example, a markers.yml can look like

  marker_name_provided:
    description: “Name slot has been set”
    slot_was_set: name
  
  marker_mood_expressed:
    description: “Unhappy or Great Mood was expressed”
    or:
      - intent: mood_unhappy
      - intent: mood_great

• Add rasa marker upload command to upload markers to the Rasa Pro Services. Usage: rasa marker upload --config=<path-to-config-file> -d=<path-to-domain-file> -rasa-pro-services-url=<url>.

• Enhance the validation of the anonymization key in endpoints.yaml by introducing checks for required fields and duplicate IDs.

Bugfixes

• Fix running custom form validation to update required slots at form activation when prefilled slots consist only of slots that are not requested by the form.
• Anonymize rasa_events structlog key.
• Fixes issue with uploading locally trained model to a cloud rasa-plus instance where the conversation does not go as expected because slots don't get set correctly, e.g. an error is logged Tried to set non existent slot 'placeholder_slot_name'. Make sure you added all your slots to your domain file.. This is because the updated domain during the cloud upload did not get passed to the wrapped tracker store of the AuthRetryTrackerStore rasa-plus component. The fix was to add domain property and setter methods to the AuthRetryTrackerStore component.
• When using rasa studio upload, if no specific intents or entities are specified by the user, the update will now include all available intents or entities.

Improved Documentation

• Explicitly set Node.js version to 12.x in order to run Docusaurus.
• Update obselete commands in docs README.
• Correct docker image name for deploy-rasa-pro-services in docs.
• Update Compatibility Matrix.
• Implement rasa data split stories to split stories data into train/test parts.
• Updated knowledge base action docs to reflect the improvements made in knowledge base actions in Rasa Pro 3.6 version. This enhancement now allows users to query for the object attribute without the need for users to request a list of objects of a particular object type beforehand. The docs update mentions this under :::info New in 3.6 section.
• Fix dead link in Analytics documentation.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.5.12] - 2023-06-23

Rasa Pro 3.5.12 (2023-06-23)

Bugfixes

• Rich responses containing buttons with parentheses characters are now correctly parsed. Previously any characters found between the first identified pair of () in response button took precedence.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.5.11] - 2023-06-08

Rasa Pro 3.5.11 (2023-06-08)

Bugfixes

• Fix running custom form validation to update required slots at form activation when prefilled slots consist only of slots that are not requested by the form.

[3.5.10] - 2023-05-23

Rasa Pro 3.5.10 (2023-05-23)

Improved Documentation

• Added documentation for spaces alpha

[3.5.9] - 2023-05-19

Rasa Pro 3.5.9 (2023-05-19)

No significant changes.

[3.5.8] - 2023-05-12

Rasa Pro 3.5.8 (2023-05-12)

Bugfixes

• Explicitly handled BufferError exception - Local: Queue full in Kafka producer.

[3.5.7] - 2023-05-09

Rasa Pro 3.5.7 (2023-05-09)

Bugfixes

• SlotSet events will be emitted when the value set by the custom action is the same as the existing value of the slot. This was fixed for AugmentedMemoizationPolicy to work properly with truncated trackers.

  To restore the previous behaviour, the custom action can return a SlotSet only if the slot value has changed. For example,

  class CustomAction(Action):
      def name(self) -> Text:
          return "custom_action"
  
      def run(self, dispatcher: CollectingDispatcher,
              tracker: Tracker,
              domain: Dict[Text, Any]) -> List[Dict[Text, Any]]:
          # current value of the slot
          slot_value = tracker.get_slot('my_slot')
  
          # value of the entity
          # this is parsed from the user utterance
          entity_value = next(tracker.get_latest_entity_values("entity_name"), None)
  
          if slot_value != entity_value:
            return[SlotSet("my_slot", entity_value)]

[3.5.6] - 2023-04-28

Rasa Pro 3.5.6 (2023-04-28)

Bugfixes

• Addresses Regular Expression Denial of Service vulnerability in slack connector (https://owasp.org/www-community/attacks/Regular_expression_Denial_of_Service_-_ReDoS)
• Fix parsing of RabbitMQ URL provided in endpoints.yml file to include vhost path and query parameters. Re-allows inclusion of credentials in the URL as a regression fix (this was supported in 2.x).

[3.5.5] - 2023-04-20

Rasa Pro 3.5.5 (2023-04-20)

Bugfixes

• Allow slot mapping parameter intent to accept a list of intent names (as strings), in addition to accepting an intent name as a single string.
• Fix BlockingIOError when running rasa shell on utterances with more than 5KB of text.
• Use ruamel.yaml round-trip loader in order to preserve all comments after appending assistant_id to config.yml.
• Fix AttributeError: 'NoneType' object has no attribute 'send_response' caused by retrieving tracker via GET /conversations/\{conversation_id\}/tracker endpoint when action_session_start is customized in a custom action. This was addressed by passing an instance of CollectingOutputChannel to the method retrieving the tracker from the MessageProcessor.

Improved Documentation

• Updated AWS model loading documentation to indicate what should AWS_ENDPOINT_URL environment variable be set to. Added integration test for AWS model loading.
• Updated Rasa Pro Services documentation to add KAFKA_SSL_CA_LOCATION environment variable. Allows connections over SSL to Kafka
• Added note to CLI documentation to address encoding and color issues on certain Windows terminals

Miscellaneous internal changes

Miscellaneous internal changes.

[3.5.4] - 2023-04-05

Rasa Pro 3.5.4 (2023-04-05)

Bugfixes

• Fix issue with failures while publishing events to RabbitMQ after a RabbitMQ restart. The fix consists of pinning aio-pika dependency to 8.2.3, since this issue was introduced in aio-pika v8.2.4.
• Patch redis Race Conditiion vulnerability.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.5.3] - 2023-03-30

Rasa Pro 3.5.3 (2023-03-30)

Improved Documentation

• Add new Rasa Pro page in docs, together with minimal content changes.

[3.5.2] - 2023-03-30

Rasa Pro 3.5.2 (2023-03-30)

Improvements

• Add a self-reference of the synonym in the EntitySynonymMapper to handle entities extracted in a casing different to synonym case. (For example if a synonym austria is added, entities extracted with any alternate casing of the synonym will also be mapped to austria). It addresses ATO-616

Bugfixes

• Make custom actions inheriting from rasa-sdk FormValidationAction parent class an exception of the selective_domain rule and always send them domain.
• Fix 2 issues detected with the HTTP API:
  • The GET /conversations/\{conversation_id\}/tracker endpoint was not returning the tracker with all sessions when include_events query parameter was set to ALL. The fix constituted in using TrackerStore.retrieve_full_tracker method instead of TrackerStore.retrieve method in the function handling the GET /conversations/\{conversation_id\}/tracker endpoint. Implemented or updated this method across all tracker store subclasses.
  • The GET /conversations/\{conversation_id\}/story endpoint was not returning all the stories for all sessions when all_sessions query parameter was set to true. The fix constituted in using all events of the tracker to be converted in stories instead of only the applied_events.

Improved Documentation

• Add documentation for secrets managers.

[3.5.1] - 2023-03-24

Rasa Pro 3.5.1 (2023-03-24)

Bugfixes

• Fixes training DIETCLassifier on the GPU.

  A deterministic GPU implementation of SparseTensorDenseMatmulOp is not currently available

Improved Documentation

• Updated Test your assistant section to describe the new end-to-end testing feature. Also updated CLI and telemetry reference docs.
• Update Compatibility Matrix.

[3.5.0] - 2023-03-21

Rasa Pro 3.5.0 (2023-03-21)

Features

• Add a new required key (assistant_id) to config.yml to uniquely identify assistants in deployment. The assistant identifier is extracted from the model metadata and added to the metadata of all dialogue events. Re-training will be required to include the assistant id in the event metadata.

  If the assistant identifier is missing from the config.yml or the default identifier value is not replaced, a random identifier is generated during each training.

  An assistant running without an identifier will issue a warning that dialogue events without identifier metadata will be streamed to the event broker.

• End-to-end testing is an enhanced and comprehensive CLI-based testing tool that allows you to test conversation scenarios with different pre-configured contexts, execute custom actions, verify response texts or names, and assert when slots are filled. It is available ysing the new rasa test e2e command.

• You can now store your assistant's secrets in an external credentials manager. In this release, Rasa Pro currently supports credentials manager for the Tracker Store with HashiCorp Vault.

Improvements

• Add capability to send compressed body in HTTP request to action server. Use COMPRESS_ACTION_SERVER_REQUEST=True to turn the feature on.

Bugfixes

• Address potentially missing events with Pika consumer due to weak references on asynchronous tasks, as specifcied in Python official documentation.
• Sets a global seed for numpy, TensorFlow, keras, Python and CuDNN, to ensure consistent random number generation.

Improved Documentation

• Clarify in the docs, how rules are designed and how to use this behaviour to abort a rule

Miscellaneous internal changes

Miscellaneous internal changes.

[3.4.14] - 2023-06-08

Rasa Pro 3.4.14 (2023-06-08)

Bugfixes

• Fix running custom form validation to update required slots at form activation when prefilled slots consist only of slots that are not requested by the form.

[3.4.13] - 2023-05-19

Rasa Pro 3.4.13 (2023-05-19)

No significant changes.

[3.4.12] - 2023-05-12

Rasa Pro 3.4.12 (2023-05-12)

Bugfixes

• Explicitly handled BufferError exception - Local: Queue full in Kafka producer.

[3.4.11] - 2023-05-09

Rasa Pro 3.4.11 (2023-05-09)

Bugfixes

• Fix parsing of RabbitMQ URL provided in endpoints.yml file to include vhost path and query parameters. Re-allows inclusion of credentials in the URL as a regression fix (this was supported in 2.x).

• SlotSet events will be emitted when the value set by the custom action is the same as the existing value of the slot. This was fixed for AugmentedMemoizationPolicy to work properly with truncated trackers.

  To restore the previous behaviour, the custom action can return a SlotSet only if the slot value has changed. For example,

  class CustomAction(Action):
      def name(self) -> Text:
          return "custom_action"
  
      def run(self, dispatcher: CollectingDispatcher,
              tracker: Tracker,
              domain: Dict[Text, Any]) -> List[Dict[Text, Any]]:
          # current value of the slot
          slot_value = tracker.get_slot('my_slot')
  
          # value of the entity
          # this is parsed from the user utterance
          entity_value = next(tracker.get_latest_entity_values("entity_name"), None)
  
          if slot_value != entity_value:
            return[SlotSet("my_slot", entity_value)]

Miscellaneous internal changes

Miscellaneous internal changes.

[3.4.10] - 2023-04-17

Rasa Pro 3.4.10 (2023-04-17)

Miscellaneous internal changes

Miscellaneous internal changes.

[3.4.9] - 2023-04-05

Miscellaneous internal changes

Miscellaneous internal changes.

[3.4.8] - 2023-04-03

Rasa Pro 3.4.8 (2023-04-03)

Bugfixes

• Fix issue with failures while publishing events to RabbitMQ after a RabbitMQ restart. The fix consists of pinning aio-pika dependency to 8.2.3, since this issue was introduced in aio-pika v8.2.4.

[3.4.7] - 2023-03-30

Rasa Pro 3.4.7 (2023-03-30)

Improvements

• Add a self-reference of the synonym in the EntitySynonymMapper to handle entities extracted in a casing different to synonym case. (For example if a synonym austria is added, entities extracted with any alternate casing of the synonym will also be mapped to austria). It addresses ATO-616

Bugfixes

• Fix 2 issues detected with the HTTP API:
  • The GET /conversations/\{conversation_id\}/tracker endpoint was not returning the tracker with all sessions when include_events query parameter was set to ALL. The fix constituted in using TrackerStore.retrieve_full_tracker method instead of TrackerStore.retrieve method in the function handling the GET /conversations/\{conversation_id\}/tracker endpoint. Implemented or updated this method across all tracker store subclasses.
  • The GET /conversations/\{conversation_id\}/story endpoint was not returning all the stories for all sessions when all_sessions query parameter was set to true. The fix constituted in using all events of the tracker to be converted in stories instead of only the applied_events.
• Make custom actions inheriting from rasa-sdk FormValidationAction parent class an exception of the selective_domain rule and always send them domain.

[3.4.6] - 2023-03-16

Rasa Pro 3.4.6 (2023-03-16)

Bugfixes

• Fixes CountVectorFeaturizer to train when min_df != 1.

[3.4.5] - 2023-03-09

Rasa Pro 3.4.5 (2023-03-09)

Bugfixes

• Check unresolved slots before initiating model training.
• Fixes the bug when a slot (with from_intent mapping which contains no input for intent parameter) will no longer fill for any intent that is not under the not_intent parameter.
• Fix validation metrics calculation when batch_size is dynamic.

Improved Documentation

• Add a link to an existing docs section on how to test the audio channel on localhost.

[3.4.4] - 2023-02-17

Rasa Pro 3.4.4 (2023-02-17)

Improvements

• Add capability to send compressed body in HTTP request to action server. Use COMPRESS_ACTION_SERVER_REQUEST=True to turn the feature on.

Bugfixes

• Fix the error which resulted during merging multiple domain files where at least one of them contains custom actions that explicitly need send_domain set as True in the domain.

[3.4.3] - 2023-02-14

Rasa Pro 3.4.3 (2023-02-14)

Improvements

• Add support for custom RulePolicy.
• Add capability to select which custom actions should receive domain when they are invoked.

Bugfixes

• Fix calling the form validation action twice for the same user message triggering a form.
• Fix conditional response does not check other conditions if first condition matches.

Improved Documentation

• Add section in tracker store docs to document the fallback tracker store mechanism.

[3.4.2] - 2023-01-27

Rasa Pro 3.4.2 (2023-01-27)

Bugfixes

• Decision to publish docs should not consider next major and minor alpha release versions.

• Exit training/running Rasa model when SpaCy runtime version is not compatible with the specified SpaCy model version.

• The new custom logging feature was not working due to small syntax issue in the argparse level.

  Previously, the file name passed using the argument --logging-config-file was never retrieved so it never creates the new custom config file with the desired formatting.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.4.1] - 2023-01-19

Rasa Pro 3.4.1 (2023-01-19)

Bugfixes

• Changed categorical slot comparison to be case insensitive.
• Exit training when transformer_size is not divisible by the number_of_attention_heads parameter and update the transformer documentations.

Improved Documentation

• Update compatibility matrix between Rasa-plus and Rasa Pro services.

[3.4.0] - 2022-12-14

Rasa Pro 3.4.0 (2022-12-14)

Features

• Add metadata to Websocket channel. Messages can now include a metadata object which will be included as metadata to Rasa. The metadata can be supplied on a user configurable key with the metadata_key setting in the socketio section of the credentials.yml.
• Use a new IVR Channel to connect your assistant to AudioCodes VoiceAI Connect.

Improvements

• Added ./docker/Dockerfile_pretrained_embeddings_spacy_it to include Spacy's Italian pre-trained model it_core_news_md.
• Replace kafka-python dependency with confluent-kafka async Producer API.
• Add support for Python 3.10 version.
• Added CLI option --logging-config-file to enable configuration of custom logs formatting.

Bugfixes

• Implements a new CLI option --jwt-private-key required to have complete support for asymmetric algorithms as specified originally in the docs.

Improved Documentation

• Clarify in the documentation how to write testing stories if a user presses a button with payload.
• Clarify prioritisation of used slot asking option in forms in documentation.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.3.8] - 2023-04-06

Miscellaneous internal changes

Miscellaneous internal changes.

[3.3.7] - 2023-03-31

Improvements

• Add a self-reference of the synonym in the EntitySynonymMapper to handle entities extracted in a casing different to synonym case. (For example if a synonym austria is added, entities extracted with any alternate casing of the synonym will also be mapped to austria). It addresses ATO-616

Bugfixes

• Fix issue with failures while publishing events to RabbitMQ after a RabbitMQ restart. The fix consists of pinning aio-pika dependency to 8.2.3, since this issue was introduced in aio-pika v8.2.4.
• Fix 2 issues detected with the HTTP API:
  • The GET /conversations/\{conversation_id\}/tracker endpoint was not returning the tracker with all sessions when include_events query parameter was set to ALL. The fix constituted in using TrackerStore.retrieve_full_tracker method instead of TrackerStore.retrieve method in the function handling the GET /conversations/\{conversation_id\}/tracker endpoint. Implemented or updated this method across all tracker store subclasses.
  • The GET /conversations/\{conversation_id\}/story endpoint was not returning all the stories for all sessions when all_sessions query parameter was set to true. The fix constituted in using all events of the tracker to be converted in stories instead of only the applied_events.

[3.3.6] - 2023-03-09

Rasa Pro 3.3.6 (2023-03-09)

Bugfixes

• Fixes the bug when a slot (with from_intent mapping which contains no input for intent parameter) will no longer fill for any intent that is not under the not_intent parameter.
• Fix validation metrics calculation when batch_size is dynamic.

[3.3.5] - 2023-02-21

No significant changes.

[3.3.4] - 2023-02-14

Rasa Pro 3.3.4 (2023-02-14)

Improvements

• Add capability to send compressed body in HTTP request to action server. Use COMPRESS_ACTION_SERVER_REQUEST=True to turn the feature on.
• Add support for custom RulePolicy.

[3.3.3] - 2022-12-01

Bugfixes

• Bypass Windows path length restrictions upon saving and reading a model archive in rasa.engine.storage.LocalModelStorage.

Improvements

• Updated tensorflow to 2.8.4.

[3.3.2] - 2022-11-30

Improvements

• Added support for camembert french bert model

Bugfixes

• Fixes RuntimeWarning: coroutine 'Bot.set_webhook' was never awaited issue encountered when starting the rasa server, which caused the Telegram bot to be unresponsive.

Improved Documentation

• The documentation was updated for Buttons using messages that start with '/'. Previously, it wrongly stated that messages with '/' bypass NLU, which is not the case.
• Add documentation for Rasa Pro Services upgrades.

[3.3.1] - 2022-11-09

Improved Documentation

• Add docs on how to set up additional data lakes for Rasa Pro analytics pipeline.
• Update migration guide and form docs with prescriptive recommendation on how to implement dynamic forms with custom slot mappings.

Improvements

• Updated numpy and scikit learn version to fix vulnerabilities of those dependencies

[3.3.0] - 2022-10-24

Features

• Tracing capabilities for your Rasa Pro assistant. Distributed tracing tracks requests as they flow through a distributed system (in this case: a Rasa assistant), sending data about the requests to a tracing backend which collects all trace data and enables inspecting it. With this version of the Tracing feature, Rasa Pro supports OpenTelemetry.
• ConcurrentRedisLockStore is a new lock store that uses Redis as a persistence layer and is safe for use with multiple Rasa server replicas.

Improvements

• Added option --offset-timestamps-by-seconds to offset the timestamp of events when using rasa export
• Rasa supports native installations on Apple Silicon (M1 / M2). Please follow the installation instructions and take a look at the limitations.
• Caching Message and Features fingerprints unless they are altered, saving up to 2/3 of fingerprinting time in our tests.
• Added package versions of component dependencies as an additional part of fingerprinting calculation. Upgrading an dependency will thus lead to a retraining of the component in the future. Also, by changing fingerprint calculation, the next training after this change will be a complete retraining.
• Export events continuously rather than loading all events in memory first when using rasa export. Events will be streamed right from the start rather than loading all events first and pushing them to the broker afterwards.
• Adds new dependency pluggy, with which it was possible to implement new plugin functionality. This plugin manager enables the extension and/or enhancement of the Rasa command line interface with functionality made available in the rasa-plus package.

Bugfixes

• Fix BlockingIOError when running rasa interactive, after the upgrade of prompt-toolkit dependency.
• Fixes a bug that lead to initial slot values being incorporated into all rules by default, thus breaking most rules when the slot value changed from its initial value
• Made logistic regression classifier output a proper intent ranking and made ranking length configurable

Deprecations and Removals

• Remove code related to Rasa X local mode as it is deprecated and scheduled for removal.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.2.13] - 2023-03-09

Rasa 3.2.13 (2023-03-09)

Bugfixes

• Fix validation metrics calculation when batch_size is dynamic.
• Fixes the bug when a slot (with from_intent mapping which contains no input for intent parameter) will no longer fill for any intent that is not under the not_intent parameter.

[3.2.12] - 2023-02-21

Features

• Add metadata to Websocket channel. Messages can now include a metadata object which will be included as metadata to Rasa. The metadata can be supplied on a user configurable key with the metadata_key setting in the socketio section of the credentials.yml.

Improvements

• Add capability to send compressed body in HTTP request to action server. Use COMPRESS_ACTION_SERVER_REQUEST=True to turn the feature on.

Bugfixes

• Decision to publish docs should not consider next major and minor alpha release versions.

[3.2.11] - 2022-12-05

Improvements

• Caching Message and Features fingerprints unless they are altered, saving up to 2/3 of fingerprinting time in our tests.

Bugfixes

• Implements a new CLI option --jwt-private-key required to have complete support for asymmetric algorithms as specified originally in the docs.

[3.2.10] - 2022-09-29

Bugfixes

• Fixes scenarios in which a slot with from_trigger_intent mapping that specifies an active_loop condition was being filled despite that active_loop not being activated. In addition, fixes scenario in which a slot with from_trigger_intent mapping without a specified active_loop mapping condition is only filled if the form gets activated. Removes unnecessary validation warning that a slot with from_trigger_intent and a mapping condition should be included in the form's required_slots.
• Fixed a bug where DIETClassier crashed during training when both masked language modelling and evaluation during training were used.

Improved Documentation

• Rasa SDK documentation lives now in Rasa Open Source documentation under the Rasa SDK category.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.2.9] - 2022-09-09

Yanked.

[3.2.8] - 2022-09-08

Bugfixes

• Fix bug where KeywordIntentClassifier overrides preceding intent classifiers' predictions although the KeyWordIntentClassifier was not matching any keywords.

[3.2.7] - 2022-08-31

Improvements

• Improve rasa data validate command so that it uses custom importers when they are defined in config file.

Bugfixes

• Re-instates the REST channel metadata feature. Metadata can be provided on the metadata key.

[3.2.6] - 2022-08-12

Bugfixes

• This fix makes sure that when a Domain object is loaded from multiple files where one file specifies a custom session config and the rest do not, the default session configuration does not override the custom session config.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.2.5] - 2022-08-05

Bugfixes

• Fix KeyError which resulted when action_two_stage_fallback got executed in a project whose domain contained slot mappings.
• Fixes regression in which slot mappings were prioritized according to reverse order as they were listed in the domain, instead of in order from first to last, as was implicitly expected in 2.x. Clarifies this implicit priority order in the docs.
• Enables the dispatching of bot messages returned as events by slot validation actions.

[3.2.4] - 2022-07-21

Bugfixes

• Added session_config key as valid domain key during domain loading from directory containing a separate domain file with session configuration.
• Run default action action_extract_slots after a custom action returns a UserUttered event to fill any applicable slots.
• Handle the case when an EndpointConfig object is given as parameter to the AwaitableTrackerStore.create() method.

[3.2.3] - 2022-07-18

Bugfixes

• • Fixed error in creating response when slack sends retry messages. Assigning None to response.text caused TypeError: Bad body type. Expected str, got NoneType.
  • Fixed Slack triggering timeout after 3 seconds if the action execution is too slow. Running on_new_message as an asyncio background task instead of a blocking await fixes this by immediately returning a response with code 200.

• Revert change in #10295 that removed running the form validation action on activation of the form before the loop is active.

• SlotSet events will be emitted when the value set by the current user turn is the same as the existing value.

  Previously, ActionExtractSlots would not emit any SlotSet events if the new value was the same as the existing one. This caused the augmented memoization policy to lose these slot values when truncating the tracker.

[3.2.2] - 2022-07-05

Improved Documentation

• Update documentation for customizable classes such as tracker stores, event brokers and lock stores.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.2.1] - 2022-06-17

Bugfixes

• Fix failed check in rasa data validate that verifies forms in rules or stories are consistent with the domain when the rule or story contains a default action as active_loop step.

[3.2.0] - 2022-06-14

Deprecations and Removals

• NLU training data in JSON format is deprecated and will be removed in Rasa Open Source 4.0. Please use rasa data convert nlu -f yaml --data <path to NLU data> to convert your NLU JSON data to YAML format before support for NLU JSON data is removed.

Improvements

• Make TrackerStore interface methods asynchronous and supply an AwaitableTrackerstore wrapper for custom tracker stores which do not implement the methods as asynchronous.
• Added flag use_gpu to TEDPolicy and UnexpecTEDIntentPolicy that can be used to enable training on CPU even when a GPU is available.
• Add --endpoints command line parameter to rasa train parser.

Bugfixes

• The azure botframework channel now validates the incoming JSON Web Tokens (including signature).

  Previously, JWTs were not validated at all.

• rasa shell now outputs custom json unicode characters instead of \uxxxx codes

Improved Documentation

• Clarify aspects of the API spec GET /status endpoint: Correct response schema for model_id - a string, not an object.

  GET /conversations/{conversation_id}/tracker: Describe each of the enum options for include_events query parameter

  POST & PUT /conversations/{conversation_id}/tracker/eventss: Events schema added for each event type

  GET /conversations/{conversation_id}/story: Clarified the all_sessions query parameter and default behaviour.

  POST /model/test/intents : Remove JSON payload option since it is not supported

  POST /model/parse: Explain what emulation_mode is and how it affects response results

Miscellaneous internal changes

Miscellaneous internal changes.

[3.1.7] - 2022-08-30

Miscellaneous internal changes

Miscellaneous internal changes.

[3.1.6] - 2022-07-20

Bugfixes

• Run default action action_extract_slots after a custom action returns a UserUttered event to fill any applicable slots.

[3.1.5] - 2022-07-15

Bugfixes

• SlotSet events will be emitted when the value set by the current user turn is the same as the existing value.

  Previously, ActionExtractSlots would not emit any SlotSet events if the new value was the same as the existing one. This caused the augmented memoization policy to lose these slot values when truncating the tracker.

[3.1.4] - 2022-06-21 No significant changes.

Upgrade dependent libraries with security vulnerabilities (Pillow, TensorFlow, ujson).

[3.1.3] - 2022-06-17

Bugfixes

• The azure botframework channel now validates the incoming JSON Web Tokens (including signature).

  Previously, JWTs were not validated at all.

• Backports fix for failed check in rasa data validate that verifies forms in rules or stories are consistent with the domain when the rule or story contains a default action as active_loop step.

[3.1.2] - 2022-06-08

Miscellaneous internal changes

Miscellaneous internal changes.

[3.1.1] - 2022-06-03

Bugfixes

• Remove warning for Rasa X localmode not being supported when the --production flag is present.
• Pin requirement for scipy<1.8.0 since scipy>=1.8.0 is not backward compatible with scipy<1.8.0 and additionally requires Python>=3.8, while Rasa supports Python 3.7 as well.
• Fix the extraction of values for slots with mapping conditions from trigger intents that activate a form, which was possible in 2.x.

[3.1.0] - 2022-03-25

Features

• Add configuration options (via env variables) for library logging.

• Support other recipe types.

  This pull request also adds support for graph recipes, see details at https://rasa.com/docs/rasa/model-configuration and check Graph Recipe page.

  Graph recipe is a raw format for specifying executed graph directly. This is useful if you need a more powerful way to specify your model creation.

• Added optional ssl_keyfile, ssl_certfile, and ssl_ca_certs parameters to the Redis tracker store.

• Added LogisticRegressionClassifier to the NLU classifiers.

  This model is lightweight and might help in early prototyping. The training times typically decrease substantially, but the accuracy might be a bit lower too.

• Added support for Python 3.9.

Improvements

• Bump TensorFlow version to 2.7.

  caution

  We can't guarantee the exact same output and hence model performance if your configuration uses LanguageModelFeaturizer. This applies to the case where the model is re-trained with the new rasa open source version without changing the configuration, random seeds, and data as well as to the case where a model trained with a previous version of rasa open source is loaded with this new version for inference.

  We suggest training a new model if you are upgrading to this version of Rasa Open Source.

• Make rasa data validate check for duplicated intents, forms, responses and slots when using domains split between multiple files.

• Add an influence_conversation flag to entites to provide a shorthand for ignoring an entity for all intents.

• Add --request-timeout command line argument to rasa shell, allowing users to configure the time a request can take before it's terminated.

Bugfixes

• Validate regular expressions in nlu training data configuration.

• Unset the default values for num_threads and finetuning_epoch_fraction to None in order to fix cases when CLI defaults override the data from config.

• Update rasa data validate to not fail when active_loop is null

• Fixes Domain loading when domain config uses multiple yml files.

  Previously not all configures attributes were necessarily known when merging Domains, and in the case of entities were not being properly assigned to intents.

• Fix max_history truncation in AugmentedMemoizationPolicy to preserve the most recent UserUttered event. Previously, AugmentedMemoizationPolicy failed to predict next action after long sequences of actions (longer than max_history) because the policy did not have access to the most recent user message.

• Add RASA_ENVIRONMENT header in Kafka only if the environmental variable is set.

• Merge domain entities as lists of dicts, not lists of lists to support entity roles and groups across multiple domains.

• Add an option to specify --domain for rasa test nlu CLI command.

Improved Documentation

• Fixed an over-indent in the Tokenizers section of the Components page of the docs.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.0.10] - 2022-03-15## [3.0.10] - 2022-03-15

Bugfixes

• Fix broken conversion from Rasa JSON NLU data to Rasa YAML NLU data.

[3.0.9] - 2022-03-11

Bugfixes

• Fix Socket IO connection issues by upgrading sanic to v21.12.

  The bug is caused by an invalid function signature and is fixed in v21.12.

  This update brings some deprecations in sanic:

  • Sanic and Blueprint may no longer have arbitrary properties attached to them
    • Fixed this by moving user defined properties to the instance.ctx object
  • Sanic and Blueprint forced to have compliant names
    • Fixed this by using string literal names instead of the module's name via __name__
  • sanic.exceptions.abort is Deprecated
    • Fixed by replacing it with sanic.exceptions.SanicException
  • sanic.response.StreamingHTTPResponse is deprecated
    • Fixed by replacing it with sanic.response.ResponseStream

• Update rasa data validate to not fail when active_loop is null

Improved Documentation

• Updated the model_confidence parameter in TEDPolicy and DIETClassifier. The linear_norm is removed as it is no longer supported.
• Added an additional step to Receiving Messages section in slack.mdx documentation. After a slack update this additional step is needed to allow direct messages to the bot.
• Backport the updated deployment docs to 3.0.x.

[3.0.8] - 2022-02-11

Improvements

• Allow single tokens in rasa end-to-end test files to be annotated with multiple entities.

  Some entity extractors (s.a. RegexEntityExtractor) can generate multiple entities from a single expression. Before this change, rasa test would fail in this case, because test stories could not be annotated correctly. New annotation option is

  stories:
    - story: Some story
      steps:
        - user: |
            cancel my [iphone][{"entity":"iphone", "value":"iphone"},{"entity":"smartphone", "value":"true"}{"entity":"mobile_service", "value":"true"}]
          intent: cancel_contract

Bugfixes

• Fixed a bug where the POST /conversations/<conversation_id>/tracker/events endpoint repeated session start events when appending events to a new tracker.

[3.0.7] - 2022-02-09

Bugfixes

• Checkpoint weights were never loaded before. Implements overwriting checkpoint weights to the final model weights after training of DIETClassifier, ResponseSelector and TEDPolicy.
• Allow arbitrary keys under each slot in the domain to allow for custom slot types.
• Fix issue with missing running event loop in MainThread when starting Rasa Open Source for Rasa X with JWT secrets.

[3.0.6] - 2022-01-28

Deprecations and Removals

• Removed CompositionView.

Bugfixes

• Fixes a bug which was caused by DIETClassifier (ResponseSelector, SklearnIntentClassifier and CRFEntityExtractor have the same issue) trying to process message which didn't have required features. Implements removing unfeaturized messages for the above-mentioned components before training and prediction.
• Enable slots with from_entity mapping that are not part of a form's required slots to be set during active loop.
• Catch ValueError for any port values that cannot be cast to integer and re-raise as RasaException during the initialisation of SQLTrackerStore.
• Use tf.function for model prediction to improve inference speed.
• Tie prompt-toolkit to ^2.0 to fix rasa-shell.

Improved Documentation

• Update dynamic form behaviour docs section with an example on how to override required_slots in case of removal of a form required slot.

[3.0.5] - 2022-01-19

Bugfixes

• Corrects transformer_size parameter value (None by default) with a default size during loading in case ResponseSelector contains transformer layers.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.0.4] - 2021-12-22

Miscellaneous internal changes

Miscellaneous internal changes.

[3.0.3] - 2021-12-16

Bugfixes

• Copy lookup tables to train and test folds in cross validation. Before, the generated folds did not have a copy of the lookup tables from the original NLU data, so that RegexEntityExtractor could not recognize any entities during the evaluation.
• Do not print warning when subintent actions have response.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.0.2] - 2021-12-09

Bugfixes

• Update SQLAlchemy version to a compatible one in case other dependencies force a lower version.
• Fix overriding of default config with custom config containing nested dictionaries. Before, the keys of a nested dictionary in the default config that were not specified in the custom config got lost.
• Add UserWarning to alert users trying to run rasa x CLI command with rasa version 3.0 or higher that rasa-x currently doesn't support rasa 3.x.

Improved Documentation

• Added note to the slot mappings section of the migration guide to recommend checking dynamic form behavior on migrated assistants.

[3.0.1] - 2021-12-02

Bugfixes

• Fix previous slots getting filled after a restart. Previously events were searched from oldest to newest which meant we would find first occurrence of a message and use slots from thereafter. Now we use the last utterance or the restart event.

Miscellaneous internal changes

Miscellaneous internal changes.

[3.0.0] - 2021-11-23

Deprecations and Removals

• Remove backwards compatibility code with Rasa Open Source 1.x, Rasa Enterprise 0.35, and other outdated backwards compatibility code in rasa.cli.x, rasa.core.utils, rasa.model_testing, rasa.model_training and rasa.shared.core.events.

• Removed Python 3.6 support as it reaches its end of life in December 2021.

• Follow through on removing deprecation warnings for synchronous EventBroker methods.

• Follow through on deprecation warnings for policies and policy ensembles.

• Follow through on deprecation warnings for rasa.shared.data.

• Follow through on deprecation warnings for the Domain. Most importantly this will enforce the schema of the forms section in the domain file. This further includes the removal of the UnfeaturizedSlot type.

• Remove deprecated change_form_to and set_form_validation methods from DialogueStateTracker.

• Remove the support of Markdown training data format. This includes:

  • reading and writing of story files in Markdown format
  • reading and writing of NLU data in Markdown format
  • reading and writing of retrieval intent data in Markdown format
  • all the Markdown examples and tests that use Markdown

• Removed automatic renaming of deprecated action action_deactivate_form to action_deactivate_loop. action_deactivate_form will just be treated like other non-existing actions from now on.

• Remove deprecated sorted_intent_examples method from TrainingData.

• Raising RasaException instead of deprecation warning when using class_from_module_path for loading types other than classes.

• Specifying the retrieve_events_from_previous_conversation_sessions kwarg for the any TrackerStore was deprecated and has now been removed. Please use the retrieve_full_tracker() method instead.

  Deserialization of pickled trackers was deprecated and has now been removed. Rasa will perform any future save operations of trackers using json serialisation.

  Removed catch for missing (deprecated) session_date when saving trackers in DynamoTrackerStore.

• Removed the deprecated dialogue policy state featurizers: BinarySingleStateFeature and LabelTokenizerSingleStateFeaturizer.

  Removed the deprecated method encode_all_actions of SingleStateFeaturizer. Use encode_all_labels instead.

• Follow through with removing deprecated policies: FormPolicy, MappingPolicy, FallbackPolicy, TwoStageFallbackPolicy, and SklearnPolicy.

  Remove warning about default value of max_history in MemoizationPolicy. The default value is now None.

• Follow through on deprecation warnings and remove code, tests, and docs for ConveRTTokenizer, LanguageModelTokenizer and HFTransformersNLP.

• rasa.shared.nlu.training_data.message.Message method get_combined_intent_response_key has been removed. get_full_intent should now be used in its place.

• Intent IDs sent with events (to kafka and elsewhere) have been removed, intent names can be used instead (or if numerical values are needed for backwards compatibility, one can also hash the names to get previous ID values, ie. hash(intent_name) is the old ID values). Intent IDs have been removed because they were providing no extra value and integers that large were problematic for some event broker implementations.

• Remove loop argument from train method in rasa. This argument became redundant when Python 3.6 support was dropped as asyncio.run became available in Python 3.7.

• Remove template_variables and e2e arguments from get_stories method of TrainingDataImporter. This argument was used in Markdown data format and became redundant once Markdown was removed.

• weight_sparsity has been removed. Developers should replace it with connection_density in the following way: connection_density = 1-weight_sparsity.

  softmax is not available as a loss_type anymore.

  The linear_norm option has been removed as possible value for model_confidence. Please, use softmax instead.

  minibatch has been removed as a value for tensorboard_log_level, use batch instead.

  Removed deprecation warnings related to the removed component config values.

• Follow through on removing deprecation warnings raised in these modules:

  • rasa/server.py

  • rasa/core/agent.py

  • rasa/core/actions/action.py

  • rasa/core/channels/mattermost.py

  • rasa/core/nlg/generator.py

  • rasa/nlu/registry.py

• Remove deprecation warnings associated with the "number_additional_patterns" parameter of rasa.nlu.featurizers.sparse_featurizer.regex_featurizer.RegexFeaturizer. This parameter is no longer needed for incremental training.

  Remove deprecation warnings associated with the "additional_vocabulary_size" parameter of rasa.nlu.featurizers.sparse_featurizer.count_vectors_featurizer.CountVectorsFeaturizer. This parameter is no longer needed for incremental training.

  Remove deprecated functions training_states_actions_and_entities and training_states_and_actions from rasa.core.featurizers.tracker_featurizers.TrackerFeaturizer. Use training_states_labels_and_entities and training_states_and_labels instead.

• Follow through on deprecation warning for NGramFeaturizer

• The CLI commands rasa data convert config and rasa data convert responses which converted from the Rasa Open Source 1 to the Rasa Open Source 2 formats were removed. Please use a Rasa Open Source 2 installation to convert your training data before moving to Rasa Open Source 3.

• rasa.core.agent.Agent.visualize was removed. Please use rasa visualize or rasa.core.visualize.visualize instead.

• Removed slot auto-fill functionality, making the key invalid to use in the domain file. The auto_fill parameter was also removed from the constructor of the Slot class. In order to continue filling slots with entities of the same name, you now have to define a from_entity mapping in the slots section of the domain. To learn more about how to migrate your 2.0 assistant, please read the migration guide.

Features

• Training data version upgraded from 2.0 to 3.0 due to breaking changes to format in Rasa Open Source 3.0

• A new experimental feature called Markers has been added. Markers allow you to define points of interest in conversations as a set of conditions that need to be met. A new command rasa evaluate markers allows you to apply these conditions to your existing tracker stores and outputs the points at which the conditions were satisfied.

• Rasa Open Source now uses the model configuration to build a

  directed acyclic graph. This graph describes the dependencies between the items in your model configuration and how data flows between them. This has two major benefits:

  • Rasa Open Source can use the computational graph to optimize the execution of your model. Examples for this are efficient caching of training steps or executing independent steps in parallel.
  • Rasa Open Source can represent different model architectures flexibly. As long as the graph remains acyclic Rasa Open Source can in theory pass any data to any graph component based on the model configuration without having to tie the underlying software architecture to the used model architecture.

  This change required changes to custom policies and custom NLU components. See the documentation for a detailed migration guide.

• Added explicit mechanism for slot filling that allows slots to be set and/or updated throughout the conversation. This mechanism is enabled by defining global slot mappings in the slots section of the domain file.

  In order to support this new functionality, implemented a new default action: action_extract_slots. This new action runs after each user turn and checks if any slots can be filled with information extracted from the last user message based on defined slot mappings.

  Since slot mappings were moved away from the forms section of the domain file, converted the form's required_slots to a list of slot names. In order to restrict certain mappings to a form, you can now use the conditions key in the mapping to define the applicable active_loop, like so:

  slots:
    location:
      type: text
      influence_conversation: false
      mappings:
      - type: from_entity
        entity: city
        conditions:
        - active_loop: booking_form

  To learn more about how to migrate your 2.0 assistant, please read the migration guide.

Improvements

• Updated the /status endpoint response payload, and relevant documentation, to return/reflect the updated 3.0 keys/values.

• Bump TensorFlow version to 2.6.

  This update brings some security benefits (see TensorFlow release notes for details). However, internal experiments suggest that it is also associated with increased train and inference time, as well as increased memory usage.

  You can read more about why we decided to update TensorFlow, and what the expected impact is here.

  If you experience a significant increase in train time, inference time, and/or memory usage, please let us know in the forum.

  Users can no longer set TF_DETERMINISTIC_OPS=1 if they are using GPU(s) because a tf.errors.UnimplementedError will be thrown by TensorFlow (read more here).

  caution

  This breaks backward compatibility of previously trained models. It is not possible to load models trained with previous versions of Rasa Open Source. Please re-train your assistant before trying to use this version.

• Added authentication support for connecting to external RabbitMQ servers. Currently user has to hardcode a username and a password in a URL in order to connect to an external RabbitMQ server.

• 1. Failed test stories will display full retrieval intents.

  2. Retrieval intents will be extracted during action prediction in test stories so that we won't have unnecessary mismatches anymore.

  Let's take this example story:

  - story: test story
    steps:
    - user: |
        what is your name?
      intent: chitchat/ask_name
    - action: utter_chitchat/ask_name
    - intent: bye
    - action: utter_bye

  Before:

    steps:
    - intent: chitchat   # 1) intent is not displayed in it's original form
    - action: utter_chitchat/ask_name  # predicted: utter_chitchat
                    # 2) retrieval intent is not extracted during action prediction and we have a mismatch
  
    - intent: bye  # some other fail
    - action: utter_bye # some other fail

  Both 1) and 2) problems are solved.

  Now:

    steps:
    - intent: chitchat/ask_name
    - action: utter_chitchat/ask_name
    - intent: bye  # some other fail
    - action: utter_bye # some other fail

• Added -i command line option to make RASA listen on a specific ip-address instead of any network interface

• rasa data validate now checks that forms referenced in active_loop directives are defined in the domain

• Every conversation event now includes in its metadata the ID of the model which loaded at the time it was created.

• Send indices of user message tokens along with the UserUttered event through the event broker to Rasa X.

• Added optional flag to convert intent ID hashes from integer to string in the KafkaEventBroker.

• Make it possible to use or functionality for slot_was_set events.

• Upgraded the spaCy dependency from version 3.0 to 3.1.

• Implemented fingerprint methods in these classes:

  • Event
  • Slot
  • DialogueStateTracker

• Added debug message that logs when a response condition is used.

• The naming scheme for trained models was changed. Unless you provide a --fixed-model-name to rasa train, Rasa Open Source will now generate a new model name using the schema <timestamp>-<random name>.tar.gz, e.g.

  • 20211018-094821-composite-pita.tar.gz (for a model containing a trained NLU and dialogue model)
  • nlu-20211018-094821-composite-pita.tar.gz (for a model containing only a trained NLU model but not a dialogue model)
  • core-20211018-094821-composite-pita.tar.gz (for a model containing only a trained dialogue model but no NLU model)

• Due to changes in the model architecture the behavior of rasa train --dry-run changed. The exit codes now have the following meaning:

  • 0 means that the model does not require an expensive retraining. However, the responses might still require updating by running rasa train
  • 1 means that one or multiple components require to be retrained.
  • 8 means that the --force flag was used and hence any cached results are ignored and the entire model is retrained.

• Machine learning components like DIETClassifier, ResponseSelector and TEDPolicy using a ranking_length parameter will no longer report renormalised confidences for the top predictions by default.

  A new parameter renormalize_confidences is added to these components which if set to True, renormalizes the confidences of top ranking_length number of predictions to sum up to 1. The default value is False, which means no renormalization will be applied by default. It is advised to leave it to False but if you are trying to reproduce the results from previous versions of Rasa Open Source, you can set it to True.

  Renormalization will only be applied if model_confidence=softmax is used.

Bugfixes

• Fixed validation behavior and logging output around unused intents and utterances.
• rasa test nlu --cross-validation uses autoconfiguration when no pipeline is defined instead of failing
• Update DynamoDb tracker store to correctly retrieve all sender_ids from a DynamoDb table.
• Fix for failed_test_stories.yml not printing the correct message when the extracted entity specified in a test story is incorrect.
• Fix CVE-2021-41127

Improved Documentation

• Added new docs for Markers.
• Update pip in same command which installs rasa and clarify supported version in docs.
• Update pika consumer code in Event Brokers documentation.
• Adds documentation on how to use CRFEntityExtractor with features from a dense featurizer (e.g. LanguageModelFeaturizer).
• Updated docs (Domain, Forms, Default Actions, Migration Guide, CLI) to provide more detail over the new slot mappings changes.
• Updated documentation publishing mechanisms to build one version of the documentation for each major version of Rasa Open Source, starting from 2.x upwards. Previously, we were building one version of the documentation for each minor version of Rasa Open Source, resulting in a poor user experience and high maintenance costs.

Miscellaneous internal changes

Miscellaneous internal changes.

[2.8.16] - 2021-12-09

Improvements

• The value of the RASA_ENVIRONMENT environmental variable is sent as a header in messages logged by KafkaEventBroker. This value was previously only made available by PikaEventConsumer.

Bugfixes

• Make action_metadata json serializable and make it available on the tracker. This is a backport of a fix in 3.0.0.

[2.8.15] - 2021-11-25

Bugfixes

• Validate regular expressions in nlu training data configuration.

[2.8.14] - 2021-11-18

Bugfixes

• Bump TensorFlow version to 2.6.2. We have plans to port this change to 3.x (see this issue).
• Downgrade google-auth to <2.

[2.8.13] - 2021-11-11

Bugfixes

• Fixed new intent creation in rasa interactive command. Previously, this failed with 500 from the server due to UnexpecTEDIntentPolicy trying to predict with the new intent not in domain.
• Install mitie library when preparing test runs. This step was missing before and tests were thus failing as we have many tests which rely on mitie library. Previously, make install-full was required.

Miscellaneous internal changes

Miscellaneous internal changes.

[2.8.12] - 2021-10-21

Bugfixes

• Fixed a bug where rasa test --fail-on-prediction-errors would raise a WrongPredictionException for entities which were actually predicted correctly.

  This happened in two ways:

  1. if for a user message some entities were extracted multiple times (by multiple entity extractors) but listed only once in the test story,
  2. if the order in which entities from a message were extracted didn't match the order in which they were listed in the test story.

Improved Documentation

• Improve the documentation for training TEDPolicy with data augmentation.

[2.8.11] - 2021-10-20

Bugfixes

• Updates dependency on sanic-jwt (1.5.0 -> ">=1.6.0, <1.7.0")

  This removes the need to pin the version of pyjwt as the newer version of sanic-jwt manages this properly.

[2.8.10] - 2021-10-14

Bugfixes

• Add List handling in the send_custom_json method on channels/facebook.py. Bellow are some examples that could cause en error before.

  Example 1: when the whole json is a List

  [
      {
          "blocks": {
              "type": "progression_bar",
              "text": {"text": "progression 1", "level": "1"},
          }
      },
      {"sender": {"id": "example_id"}},
  ]

  Example 2: instead of being a Dict, blocks is a List when there are 2 type keys under it

  {
      "blocks": [
          {"type": "title", "text": {"text": "Conversation progress"}},
          {
              "type": "progression_bar",
              "text": {"text": "Look how far we are...", "level": "1"},
          },
      ]
  }

• Fixed bug when using wit.ai training data to train. Training failed with an error similarly to this:

    File "./venv/lib/python3.8/site-packages/rasa/nlu/classifiers/diet_classifier.py", line 803, in train
      self.check_correct_entity_annotations(training_data)
    File "./venv/lib/python3.8/site-packages/rasa/nlu/extractors/extractor.py", line 418, in check_correct_entity_annotations
      entities_repr = [
    File "./venv/lib/python3.8/site-packages/rasa/nlu/extractors/extractor.py", line 422, in \<listcomp\>
      entity[ENTITY_ATTRIBUTE_VALUE],
  KeyError: 'value'

• Fix CVE-2021-41127

[2.8.9] - 2021-10-08

Improvements

• Bump TensorFlow version to 2.6.

  This update brings some security benefits (see TensorFlow release notes for details). However, internal experiments suggest that it is also associated with increased train and inference time, as well as increased memory usage.

  You can read more about why we decided to update TensorFlow, and what the expected impact is here.

  If you experience a significant increase in train time, inference time, and/or memory usage, please let us know in the forum.

  Users can no longer set TF_DETERMINISTIC_OPS=1 if they are using GPU(s) because a tf.errors.UnimplementedError will be thrown by TensorFlow (read more here).

  caution

  This breaks backward compatibility of previously trained models. It is not possible to load models trained with previous versions of Rasa Open Source. Please re-train your assistant before trying to use this version.

[2.8.8] - 2021-10-06

Improvements

• Added a function to display the actual text of a Token when inspecting a Message in a pipeline, making it easier to debug.

Improved Documentation

• Removing the experimental feature warning for conditional response variations from the Rasa docs. The behaviour of the feature remains unchanged.
• Updates quick install documentation with optional venv step, better pip install instructions, & M1 warning

[2.8.7] - 2021-09-20

Bugfixes

• Explicitly set the upper limit for currently compatible TensorFlow versions.

[2.8.6] - 2021-09-09

Bugfixes

• Fix rules not being applied when a featurised categorical slot has as one of its allowed values none, NoNe, None or a similar value.

[2.8.5] - 2021-09-06

Bugfixes

• AugmentedMemoizationPolicy is accelerated for large trackers
• Bump tensorflow to 2.3.4 to address security vulnerabilities

[2.8.4] - 2021-09-02

Improvements

• Increase speed of augmented lookup for AugmentedMemoizationPolicy

Bugfixes

• Fix --data being treated as if non-optional on sub-commands of rasa data convert
• Fixes bug where hide_rule_turn was defaulting to None when ActionExecuted was deserialised.

Miscellaneous internal changes

Miscellaneous internal changes.

[2.8.3] - 2021-08-19

Bugfixes

• Ignore checking that intent is in domain for E2E story utterances when running rasa data validate. Previously data validation would fail on E2E stories.

[2.8.2] - 2021-08-04

Bugfixes

• Fixes a bug which caused training of UnexpecTEDIntentPolicy to crash when end-to-end training stories were included in the training data.

  Stories with end-to-end training data will now be skipped for the training of UnexpecTEDIntentPolicy.

Improved Documentation

• Removing the experimental feature warning for the story validation tool from the rasa docs. The behaviour of the feature remains unchanged.
• Removing the experimental feature warning for entity roles and groups from the rasa docs, and from the code where it previously appeared as a print statement. The behaviour of the feature remains otherwise unchanged.

[2.8.1] - 2021-07-22

Improvements

• Add support for cafile parameter in endpoints.yaml. This will load a custom local certificate file and use it when making requests to that endpoint.

  For example:

  action_endpoint:
    url: https://localhost:5055/webhook
    cafile: ./cert.pem

  This means that requests to the action server localhost:5055 will use the certificate cert.pem located in the current working directory.

Bugfixes

• Fixes wrong overriding of epochs parameter when TEDPolicy or UnexpecTEDIntentPolicy is not loaded in finetune mode.

[2.8.0] - 2021-07-12

Deprecations and Removals

• The option model_confidence=linear_norm is deprecated and will be removed in Rasa Open Source 3.0.0.

  Rasa Open Source 2.3.0 introduced linear_norm as a possible value for model_confidence parameter in machine learning components such as DIETClassifier, ResponseSelector and TEDPolicy. Based on user feedback, we have identified multiple problems with this option. Therefore, model_confidence=linear_norm is now deprecated and will be removed in Rasa Open Source 3.0.0. If you were using model_confidence=linear_norm for any of the mentioned components, we recommend to revert it back to model_confidence=softmax and re-train the assistant. After re-training, we also recommend to re-tune the thresholds for fallback components.

• The fallback mechanism for spaCy models has now been removed in Rasa 3.0.0.

  Rasa Open Source 2.5.0 introduced support for spaCy 3.0. This introduced a breaking feature because models would no longer be manually linked. To make the transition smooth Rasa would rely on the language parameter in the config.yml to fallback to a medium spaCy model if no model was configured for the SpacyNLP component. In Rasa Open Source 3.0.0 and onwards the SpacyNLP component will require the model name (like "en_core_web_md") to be passed explicitly.

Features

• Added sasl_mechanism as an optional configurable parameters for the Kafka Producer.

• Introduces a new policy called UnexpecTEDIntentPolicy.

  UnexpecTEDIntentPolicy helps you review conversations and also allows your bot to react to unexpected user turns in conversations. It is an auxiliary policy that should only be used in conjunction with at least one other policy, as the only action that it can trigger is the special and newly introduced action_unlikely_intent action.

  The auto-configuration will include UnexpecTEDIntentPolicy in your configuration automatically, but you can also include it yourself in the policies section of the configuration:

  policies:
    - name: UnexpecTEDIntentPolicy
      epochs: 200
      max_history: 5

  As part of the feature, it also introduces:

  • IntentMaxHistoryTrackerFeaturizer to featurize the trackers for UnexpecTEDIntentPolicy.
  • MultiLabelDotProductLoss to support UnexpecTEDIntentPolicy's multi-label training objective.
  • A new default action called action_unlikely_intent.

  rasa test command has also been adapted to UnexpecTEDIntentPolicy:

  • If a test story contains action_unlikely_intent and the policy ensemble does not trigger it, this leads to a test error (wrongly predicted action) and the corresponding story will be logged in failed_test_stories.yml.
  • If the story does not contain action_unlikely_intent and Rasa Open Source does predict it then the prediction of action_unlikely_intent will be ignored for the evaluation (and hence not lead to a prediction error) but the story will be logged in a file called stories_with_warnings.yml.

  The rasa data validate command will warn if action_unlikely_intent is included in the training stories. Accordingly, YAMLStoryWriter and MarkdownStoryWriter have been updated to not dump action_unlikely_intent when writing stories to a file.

  caution

  The introduction of a new default action breaks backward compatibility of previously trained models. It is not possible to load models trained with previous versions of Rasa Open Source. Please re-train your assistant before trying to use this version.

Improvements

• Added detailed json schema validation for UserUttered, SlotSet, ActionExecuted and EntitiesAdded events both sent and received from the action server, as well as covered at high-level the validation of the rest of the 20 events. In case the events are invalid, a ValidationError will be raised.

• Users don't need to specify an additional buffer size for sparse featurizers anymore during incremental training.

  Space for new sparse features are created dynamically inside the downstream machine learning models - DIETClassifier, ResponseSelector. In other words, no extra buffer is created in advance for additional vocabulary items and space will be dynamically allocated for them inside the model.

  This means there's no need to specify additional_vocabulary_size for CountVectorsFeaturizer or number_additional_patterns for RegexFeaturizer. These parameters are now deprecated.

  Before

  pipeline:
    - name: "WhitespaceTokenizer"
    - name: "RegexFeaturizer"
      number_additional_patterns: 100
    - name: "CountVectorsFeaturizer"
      additional_vocabulary_size: {text: 100, response: 20}

  Now

  pipeline:
    - name: "WhitespaceTokenizer"
    - name: "RegexFeaturizer"
    - name: "CountVectorsFeaturizer"

  Also, all custom layers specifically built for machine learning models - RasaSequenceLayer, RasaFeatureCombiningLayer and ConcatenateSparseDenseFeatures now inherit from RasaCustomLayer so that they support flexible incremental training out of the box.

• Speed up the contradiction check of the RulePolicy by a factor of 3.

• Change the confidence score assigned by FallbackClassifier to fallback intent to be the same as the fallback threshold.

• Issue a UserWarning if a specified domain folder contains files that look like YML files but cannot be parsed successfully. Only invoked if user specifies a folder path in --domain paramater. Previously those invalid files in the specified folder were silently ignored. Does not apply to individually specified domain YAML files, e.g. --domain /some/path/domain.yml, those being invalid will still raise an exception.

Bugfixes

• Fix for unnecessary retrain and duplication of folders in the model

Miscellaneous internal changes

Miscellaneous internal changes.

[2.7.2] - 2021-08-09

Bugfixes

• Ignore checking that intent is in domain for E2E story utterances when running rasa data validate. Previously data validation would fail on E2E stories.
• Fix for unnecessary retrain and duplication of folders in the model

[2.7.1] - 2021-06-16

Bugfixes

• Best model checkpoint allows for metrics to be equal to previous best if at least one metric improves, rather than strict improvement for each metric.

• Fixes a bug where multiple plots overlap each other and are rendered incorrectly when comparing performance across multiple NLU pipelines.

• Don't evaluate entities if no entities present in test data.

  Also, catch exception in plot_paired_histogram when data is empty.

[2.7.0] - 2021-06-03

Improvements

• Changed the default config to train the RulePolicy before the TEDPolicy. This means that conflicting rule/stories will be identified before a potentially slow training of the TEDPolicy.
• Updated validator used by rasa data validate to verify that actions used in stories and rules are present in the domain and that form slots match domain slots.
• Rename plot_histogram to plot_paired_histogram and fix missing bars in the plot.
• Changed --data option type in the ``rasa data validate``` command to allow more than one path to be passed.

Bugfixes

• The file failed_test_stories.yml (generated by rasa test) now also includes the wrongly predicted entity as a comment next to the entity of a user utterance. Additionally, the comment printed next to the intent of a user utterance is printed only if the intent was wrongly predicted (irrelevantly if there was a wrongly predicted entity or not in the specific user utterance).
• Added check in PikaEventBroker constructor: if port cannot be cast to integer, raise RasaException
• Fixed bug where missing intent warnings appear when running rasa test
• Update should_retrain function to return the correct fingerprint comparison result even when there is a problem with model unpacking.
• Handle correctly Telegram edited message.

Miscellaneous internal changes

Miscellaneous internal changes.

[2.6.3] - 2021-05-28

Bugfixes

• ResponseSelector can now be trained with the transformer enabled (i.e. when a positive number_of_transformer_layers is provided) even if one doesn't specify the transformer's size. Previously, not specifying transformer_size led to an error.
• Return EntityEvaluationResult during evaluation of test stories only if parsed_message is not None.
• Ignore OSError in Sentry reporting.
• Replaced ValueError with RasaException in TED model _check_data method.
• Changed import to fix agent creation in Jupyter.

Miscellaneous internal changes

Miscellaneous internal changes.

[2.6.2] - 2021-05-18

Bugfixes

• Fixed a bug where ListSlots were filled with single items in case only one matching entity was extracted for this slot.

  Values applied to ListSlots will be converted to a List in case they aren't one.

• Fix bug with false rule conflicts

  This essentially reverts PR 8446, except for the tests. The PR is redundant due to PR 8646.

• Handle AttributeError thrown by empty slot mappings in domain form through refactoring.

• Fixed incorrect The action 'utter_\<response selector intent\>' is used in the stories, but is not a valid utterance action error when running rasa data validate with response selector responses in the domain file.

Improved Documentation

• Added a note to clarify best practice for resetting all slots after form deactivation.

Miscellaneous internal changes

Miscellaneous internal changes.

[2.6.1] - 2021-05-11

Bugfixes

• Made SchemaError message available to validator so that the reason why reason schema validation fails during rasa data validate is displayed when response text value is null. Added warning message when deprecated MappingPolicy format is used in the domain.

• When there are multiple entities in a user message, they will get sorted when creating a representation of the current dialogue state.

  Previously, the ordering was random, leading to inconsistent state representations. This would sometimes lead to memoization policies failing to recall a memorised action.

[2.6.0] - 2021-05-06

Deprecations and Removals

• In forms, the keyword required_slots should always precede the definition of slot mappings and the lack of it is deprecated. Please see the migration guide for more information.
• rasa.data.get_test_directory, rasa.data.get_core_nlu_directories, and rasa.shared.nlu.training_data.training_data.TrainingData::get_core_nlu_directories are deprecated and will be removed in Rasa Open Source 3.0.0.
• Update the minimum compatible model version to "2.6.0". This means all models trained with an earlier version will have to be retrained.

Features

• Feature enhancement enabling JWT authentication for the Socket.IO channel. Users can define jwt_key and jwt_method as parameters in their credentials file for authentication.

• Allows a Rasa bot to be connected to a Twilio Voice channel. More details in the Twilio Voice docs

• Conditional response variations are supported in the domain.yml without requiring users to write custom actions code.

  A condition can be a list of slot-value mapping constraints.

Improvements

• Added an optional ignored_intents parameter in forms.

  • To use it, add the ignored_intents parameter in your domain.yml file after the forms name and provide a list of intents to ignore. Please see Forms for more information.
  • This can be used in case the user never wants to fill any slots of a form with the specified intent, e.g. chitchat.

• Add function to carry max_history to featurizer

• Improved the machine learning models' codebase by factoring out shared feature-processing logic into three custom layer classes:

  • ConcatenateSparseDenseFeatures combines multiple sparse and dense feature tensors into one.
  • RasaFeatureCombiningLayer additionally combines sequence-level and sentence-level features.
  • RasaSequenceLayer is used for attributes with sequence-level features; it additionally embeds the combined features with a transformer and facilitates masked language modeling.

• Added the following usability improvements with respect to entities getting extracted multiple times:

  • Added warnings for competing entity extractors at training time and for overlapping entities at inference time
  • Improved docs to help users handle overlapping entity problems.

• Replace weight_sparsity with connection_density in all transformer-based models and add guarantees about internal layers.

  We rename DenseWithSparseWeights into RandomlyConnectedDense, and guarantee that even at density zero the output is dense and every input is connected to at least one output. The former weight_sparsity parameter of DIET, TED, and the ResponseSelector, is now roughly equivalent to 1 - connection_density, except at very low densities (high sparsities).

  All layers and components that used to have a sparsity argument (Ffnn, TransformerRasaModel, MultiHeadAttention, TransformerEncoderLayer, TransformerEncoder) now have a density argument instead.

• Rasa test now prints a warning if the test stories contain bot utterances that are not part of the domain.

• Updated asyncio.Task.all_tasks to asyncio.all_tasks, with a fallback for python 3.6, which raises an AttributeError for asyncio.all_tasks. This removes the deprecation warning for the Task.all_tasks usage.

• Change variable name from i to array_2D

• Implement a new interface run_inference inside RasaModel which performs batch inferencing through tensorflow models.

  rasa_predict inside RasaModel has been made a private method now by changing it to _rasa_predict.

Bugfixes

• Fixed a bug for plotting trackers with non-ascii texts during interactive training by enforcing utf-8 encoding
• Fix masked language modeling in DIET to only apply masking to token-level (sequence-level) features. Previously, masking was applied to both token-level and sentence-level features.
• Make it possible to use null entities in stories.
• Introduce a skip_validation flag in order to speed up reading YAML files that were already validated.
• Fixed a bug in interactive training that lead to crashes for long Chinese, Japanese, or Korean user or bot utterances.

[2.5.2] - 2021-06-16

Features

• Added sasl_mechanism as an optional configurable parameters for the Kafka Producer.

[2.5.1] - 2021-04-28

Bugfixes

• Fixed prediction for rules with multiple entities.
• Mitigated Matplotlib backend issue using lazy configuration and added a more explicit error message to guide users.

[2.5.0] - 2021-04-12

Deprecations and Removals

• The following import abbreviations were removed:
  • rasa.core.train: Please use rasa.core.train.train instead.
  • rasa.core.visualize: Please use rasa.core.visualize.visualize instead.
  • rasa.nlu.train: Please use rasa.nlu.train.train instead.
  • rasa.nlu.test: Please use rasa.nlu.test.run_evaluation instead.
  • rasa.nlu.cross_validate: Please use rasa.nlu.test.cross_validate instead.

Features

• Upgraded Rasa to be compatible with spaCy 3.0.

  This means that we can support more features for more languages but there are also a few changes.

  SpaCy 3.0 deprecated the spacy link \<language model\> command so that means that from now on the full model name needs to be used in the config.yml file.

  Before

  Before you could run spacy link en en_core_web_md and then we would be able to pick up the correct model from the language parameter.

  language: en
  
  pipeline:
     - name: SpacyNLP

  Now

  This behavior will be deprecated and instead you'll want to be explicit in config.yml.

  language: en
  
  pipeline:
     - name: SpacyNLP
       model: en_core_web_md

  Fallback

  To make the transition easier, Rasa will try to fall back to a medium spaCy model when-ever a compatible language is configured for the entire pipeline in config.yml even if you don't specify a model. This fallback behavior is temporary and will be deprecated in Rasa 3.0.0.

  We've updated our docs to reflect these changes. All examples now show a direct link to the correct spaCy model. We've also added a warning to the SpaCyNLP docs that explains the fallback behavior.

Improvements

• Improved CLI startup time.

• Add augmentation and num_threads arguments to API POST /model/train

  Fix boolean casting issue for force_training and save_to_default_model_directory arguments

• Add minimum compatible version to --version command

• Updated warning for unexpected slot events during prediction time to Rasa Open Source 2.0 YAML training data format.

• Hide dialogue turns predicted by RulePolicy in the tracker states for ML-only policies like TEDPolicy if those dialogue turns only appear as rules in the training data and do not appear in stories.

  Add set_shared_policy_states(...) method to all policies. This method sets _rule_only_data dict with keys:

  • rule_only_slots: Slot names, which only occur in rules but not in stories.
  • rule_only_loops: Loop names, which only occur in rules but not in stories.

  This information is needed for correct featurization to hide dialogue turns that appear only in rules.

• Faster reading of YAML NLU training data files.

• Added partition_by_sender flag to Kafka Producer to optionally associate events with Kafka partition based on sender_id.

Bugfixes

• Fixed the 'loading model' message which was logged twice when using rasa run.

• Change training data validation to only count nlu training examples.

• Rule tracker states no longer include the initial value of slots. Rules now only require slot values when explicitly stated in the rule.

• rasa test, rasa test core and rasa test nlu no longer show temporary paths in case there are issues in the test files.

• Resolved memory problems with dense features and CRFEntityExtractor

• Handle empty intent and entity mapping in the domain.

  There is now an InvalidDomain exception raised if in the domain.yml file there are empty intent or entity mappings. An example of empty intent and entity mappings is the following :

  intents:
    - greet:
    - goodbye:
  
  entities:
    - cuisine:
    - number:

• Fixed a bug in a form where slot mapping doesn't work if the predicted intent name is substring for another intent name.

• Fixes bug where stories could not be retrieved if entities had no start or end.

• Catch ChannelNotFoundEntity exception coming from the pika broker and raise as ConnectionException.

• Fix bug with NoReturn throwing an exception in Python 3.7.0 when running rasa train

• Throw RasaException instead of ValueError in situations when environment variables specified in YAML cannot be expanded.

• Updated python-engineio version for compatibility with python-socketio

Miscellaneous internal changes

Miscellaneous internal changes.

[2.4.3] - 2021-03-26

Bugfixes

• Fixes bug where stories could not be retrieved if entities had no start or end.

[2.4.2] - 2021-03-25

Bugfixes

• Fix UnicodeException in is_key_in_yaml.
• Fixed the bug that events from previous conversation sessions would be re-saved in the SQLTrackerStore or MongoTrackerStore when retrieve_events_from_previous_conversation_sessions was true.

[2.4.1] - 2021-03-23

Bugfixes

• Fix TEDPolicy training e2e entities when no entities are present in the stories but there are entities in the domain.
• Fixed missing model configuration file validation.
• In Rasa 2.4.0, support for using template in utter_message when handling a custom action was wrongly deprecated. Both template and response are now supported, though note that template will be deprecated at Rasa 3.0.0.

[2.4.0] - 2021-03-11

Deprecations and Removals

• NLG Server

  • Changed request format to send response as well as template as a field. The template field will be removed in Rasa Open Source 3.0.0.

  rasa.core.agent

  • The terminology template is deprecated and replaced by response. Support for template from the NLG response will be removed in Rasa Open Source 3.0.0. Please see here for more details.

  rasa.core.nlg.generator

  • generate() now takes in utter_action as a parameter.
  • The terminology template is deprecated and replaced by response. Support for template in the NaturalLanguageGenerator will be removed in Rasa Open Source 3.0.0.

  rasa.shared.core.domain

  • The property templates is deprecated. Use responses instead. It will be removed in Rasa Open Source 3.0.0.
  • retrieval_intent_templates will be removed in Rasa Open Source 3.0.0. Please use retrieval_intent_responses instead.
  • is_retrieval_intent_template will be removed in Rasa Open Source 3.0.0. Please use is_retrieval_intent_response instead.
  • check_missing_templates will be removed in Rasa Open Source 3.0.0. Please use check_missing_responses instead.

  Response Selector

  • The field template_name will be deprecated in Rasa Open Source 3.0.0. Please use utter_action instead. Please see here for more details.
  • The field response_templates will be deprecated in Rasa Open Source 3.0.0. Please use responses instead. Please see here for more details.

Improvements

• The following endpoints now require the existence of the conversation for the specified conversation ID, raising an exception and returning a 404 status code.

  • GET /conversations/\<conversation_id:path\>/story

  • POST /conversations/\<conversation_id:path\>/execute

  • POST /conversations/\<conversation_id:path\>/predict

• Simplify our training by overwriting train_step instead of fit for our custom models.

  This allows us to use the build-in callbacks from Keras, such as the Tensorboard Callback, which offers more functionality compared to what we had before.

  warning

  If you want to use Tensorboard for DIETClassifier, ResponseSelector, or TEDPolicy and log metrics after every (mini)batch, please use 'batch' instead of 'minibatch' as 'tensorboard_log_level'.

• When TED is configured to extract entities rasa test now evaluates them against the labels in the test stories. Results are saved in /results along with the results for the NLU components that extract entities.

• We're now running integration tests for Rasa Open Source, with initial coverage for SQLTrackerStore (with PostgreSQL), RedisLockStore (with Redis) and PikaEventBroker (with RabbitMQ). The integration tests are now part of our CI, and can also be ran locally using make test-integration (see Rasa Open Source README for more information).

• Allow tests to be located anywhere, not just in tests directory.

• Model configuration files are now validated whether they match the expected schema.

• Speed up YAMLStoryReader.is_key_in_yaml function by making it to check if key is in YAML without actually parsing the text file.

• Speed up YAML parsing by reusing parsers, making the process of environment variable interpolation optional, and by not adding duplicating implicit resolvers and YAML constructors to ruamel.yaml

• Drastically improved finger printing time for large story graphs

• Remove console logging of conversation level F1-score and precision since these calculations were not meaningful.

  Add conversation level accuracy to core policy results logged to file in story_report.json after running rasa test core or rasa test.

• Improved the lock store debug log message when the process has to queue because other messages have to be processed before this item.

Bugfixes

• Fixed the bug that OR statements in stories would break the check whether a model needs to be retrained

• Update the spec of POST /model/test/intents and add tests for cases when JSON is provided.

  Fix the incorrect temporary file extension for the data that gets extracted from the payload provided in the body of POST /model/test/intents request.

• Fix for the cli command rasa data convert config when migrating Mapping Policy and no rules.

  Making rasa data convert config migrate correctly the Mapping Policy when no rules are available. It updates the config.yml file by removing the MappingPolicy and adding the RulePolicy instead. Also, it creates the data/rules.yml file even if empty in the case of no available rules.

• Allow to have slots with values that result to a dictionary under the key slot_was_set (in stories.yml file).

  An example would be to have the following story step in stories.yml:

  - slot_was_set:
      - some_slot:
          some_key: 'some_value'
          other_key: 'other_value'

  This would be allowed if the some_slot is also set accordingly in the domain.yml with type any.

• Update the fingerprinting function to recognize changes in lookup files.

• Fixed a bug when interpolating environment variables in YAML files which included $ in their value. This led to the following stack trace:

  ValueError: Error when trying to expand the environment variables in '${PASSWORD}'. Please make sure to also set these environment variables: '['$qwerty']'.
  (13 additional frame(s) were not displayed)
  ...
    File "rasa/utils/endpoints.py", line 26, in read_endpoint_config
      content = rasa.shared.utils.io.read_config_file(filename)
    File "rasa/shared/utils/io.py", line 527, in read_config_file
      content = read_yaml_file(filename)
    File "rasa/shared/utils/io.py", line 368, in read_yaml_file
      return read_yaml(read_file(filename, DEFAULT_ENCODING))
    File "rasa/shared/utils/io.py", line 349, in read_yaml
      return yaml_parser.load(content) or {}
    File "rasa/shared/utils/io.py", line 314, in env_var_constructor
      " variables: '{}'.".format(value, not_expanded)

• The REQUESTED_SLOT always belongs to the currently active form.

  Previously it was possible that after form switching, the REQUESTED_SLOT was for the previous form.

• Update the LanguageModelFeaturizer tests to reflect new default model weights for bert, and skip all bert tests with default model weights on CI, run bert tests with bert-base-uncased on CI instead.

Improved Documentation

• Update links to Sanic docs in the documentation.
• Update Rasa Playground to correctly use tracking_id when calling API methods.

Miscellaneous internal changes

Miscellaneous internal changes.

[2.3.5] - 2021-06-16

Features

• Added sasl_mechanism as an optional configurable parameters for the Kafka Producer.

Improvements

• Drastically improved finger printing time for large story graphs
• Improved the lock store debug log message when the process has to queue because other messages have to be processed before this item.

Bugfixes

• Fixed the bug that OR statements in stories would break the check whether a model needs to be retrained
• Updated python-engineio dependency version for compatibility with python-socketio.

Improved Documentation

• Update links to Sanic docs in the documentation.

[2.3.4] - 2021-02-26

Bugfixes

• Setting model_confidence=cosine in DIETClassifier, ResponseSelector and TEDPolicy is deprecated and will no longer be available. This was introduced in Rasa Open Source version 2.3.0 but post-release experiments suggest that using cosine similarity as model's confidences can change the ranking of predicted labels which is wrong.

  model_confidence=inner is deprecated and is replaced by model_confidence=linear_norm as the former produced an unbounded range of confidences which broke the logic of assistants in various other places.

  We encourage you to try model_confidence=linear_norm which will produce a linearly normalized version of dot product similarities with each value in the range [0,1]. This can be done with the following config:

  - name: DIETClassifier
    model_confidence: linear_norm
    constrain_similarities: True

  This should ease up tuning fallback thresholds as confidences for wrong predictions are better distributed across the range [0, 1].

  If you trained a model with model_confidence=cosine or model_confidence=inner setting using previous versions of Rasa Open Source, please re-train by either removing the model_confidence option from the configuration or setting it to linear_norm.

  model_confidence=cosine is removed from the configuration generated by auto-configuration.

[2.3.3] - 2021-02-25

Bugfixes

• Fixed bug where the conversation does not lock before handling a reminder event.

[2.3.2] - 2021-02-22

Bugfixes

• Fix a bug where, if a user injects an intent using the HTTP API, slot auto-filling is not performed on the entities provided.

[2.3.1] - 2021-02-17

Bugfixes

• Fixed a YAML validation error which happened when executing multiple validations concurrently. This could e.g. happen when sending concurrent requests to server endpoints which process YAML training data.

[2.3.0] - 2021-02-11

Improvements

• Expose diagnostic data for action and NLU predictions.

  Add diagnostic_data field to the Message and Prediction objects, which contain information about attention weights and other intermediate results of the inference computation. This information can be used for debugging and fine-tuning, e.g. with RasaLit.

  For examples of how to access the diagnostic data, see here.

• Using the TrainingDataImporter interface to load the data in rasa test core.

  Failed test stories are now referenced by their absolute path instead of the relative path.

• Improve error handling and Sentry tracking:

  • Raise MarkdownException when training data in Markdown format cannot be read.
  • Raise InvalidEntityFormatException error instead of json.JSONDecodeError when entity format is in valid in training data.
  • Gracefully handle empty sections in endpoint config files.
  • Introduce ConnectionException error and raise it when TrackerStore and EventBroker cannot connect to 3rd party services, instead of raising exceptions from 3rd party libraries.
  • Improve rasa.shared.utils.common.class_from_module_path function by making sure it always returns a class. The function currently raises a deprecation warning if it detects an anomaly.
  • Ignore MemoryError and asyncio.CancelledError in Sentry.
  • rasa.shared.utils.validation.validate_training_data now raises a SchemaValidationError when validation fails (this error inherits jsonschema.ValidationError, ensuring backwards compatibility).

• Allow PolicyEnsemble in cases where calling individual policy's load method returns None.

• User message metadata can now be accessed via the default slot session_started_metadata during the execution of a custom action_session_start.

  from typing import Any, Text, Dict, List
  from rasa_sdk import Action, Tracker
  from rasa_sdk.events import SlotSet, SessionStarted, ActionExecuted, EventType
  
  class ActionSessionStart(Action):
      def name(self) -> Text:
          return "action_session_start"
  
      async def run(
        self, dispatcher, tracker: Tracker, domain: Dict[Text, Any]
      ) -> List[Dict[Text, Any]]:
          metadata = tracker.get_slot("session_started_metadata")
  
          # Do something with the metadata
          print(metadata)
  
          # the session should begin with a `session_started` event and an `action_listen`
          # as a user message follows
          return [SessionStarted(), ActionExecuted("action_listen")]

• Add BILOU tagging schema for entity extraction in end-to-end TEDPolicy.

• Added two new parameters constrain_similarities and model_confidence to machine learning (ML) components - DIETClassifier, ResponseSelector and TEDPolicy.

  Setting constrain_similarities=True adds a sigmoid cross-entropy loss on all similarity values to restrict them to an approximate range in DotProductLoss. This should help the models to perform better on real world test sets. By default, the parameter is set to False to preserve the old behaviour, but users are encouraged to set it to True and re-train their assistants as it will be set to True by default from Rasa Open Source 3.0.0 onwards.

  Parameter model_confidence affects how model's confidence for each label is computed during inference. It can take three values:

  1. softmax - Similarities between input and label embeddings are post-processed with a softmax function, as a result of which confidence for all labels sum up to 1.
  2. cosine - Cosine similarity between input label embeddings. Confidence for each label will be in the range [-1,1].
  3. inner - Dot product similarity between input and label embeddings. Confidence for each label will be in an unbounded range.

  Setting model_confidence=cosine should help users tune the fallback thresholds of their assistant better. The default value is softmax to preserve the old behaviour, but we recommend using cosine as that will be the new default value from Rasa Open Source 3.0.0 onwards. The value of this option does not affect how confidences are computed for entity predictions in DIETClassifier and TEDPolicy.

  With both the above recommendations, users should configure their ML component, e.g. DIETClassifier, as

  - name: DIETClassifier
    model_confidence: cosine
    constrain_similarities: True
    ...

  Once the assistant is re-trained with the above configuration, users should also tune fallback confidence thresholds.

  Configuration option loss_type=softmax is now deprecated and will be removed in Rasa Open Source 3.0.0 . Use loss_type=cross_entropy instead.

  The default auto-configuration is changed to use constrain_similarities=True and model_confidence=cosine in ML components so that new users start with the recommended configuration.

  EDIT: Some post-release experiments revealed that using model_confidence=cosine is wrong as it can change the order of predicted labels. That's why this option was removed in Rasa Open Source version 2.3.3. model_confidence=inner is deprecated as it produces an unbounded range of confidences which can break the logic of assistants in various other places. Please use model_confidence=linear_norm which will produce a linearly normalized version of dot product similarities with each value in the range [0,1]. Please read more about this change under the notes for release 2.3.4.

• Use simple random uniform distribution of integers in negative sampling, because negative sampling with tf.while_loop and random shuffle inside creates a memory leak.

• Added support to configure exchange_name for pika event broker.

• If MaxHistoryTrackerFeaturizer is used, invert the dialogue sequence before passing it to the transformer so that the last dialogue input becomes the first one and therefore always have the same positional encoding.

Bugfixes

• Fixed an error when using the endpoint GET /conversations/\<conversation_id:path\>/story with a tracker which contained slots.

• Add the option to configure whether extracted entities should be split by comma (",") or not to TEDPolicy. Fixes crash when this parameter is accessed during extraction.

• When switching forms, the next form will always correctly ask for the first required slot.

  Before, the next form did not ask for the slot if it was the same slot as the requested slot of the previous form.

• Fix the bug when RulePolicy handling loop predictions are overwritten by e2e TEDPolicy.

• When switching forms, the next form is cleanly activated.

  Before, the next form was correctly activated, but the previous form had wrongly uttered the response that asked for the requested slot when slot validation for that slot had failed.

• Fix a bug in incremental training when passing a specific model path with the --finetune argument.

• Fix the role of unidirectional_encoder in TED. This parameter is only applied to transformers for text, action_text and label_action_text.

Miscellaneous internal changes

Miscellaneous internal changes.

[2.2.10] - 2021-02-08

Improvements

• Updated error message when using incompatible model versions.

Bugfixes

• Limit numpy version to < 1.2 as tensorflow is not compatible with numpy versions >= 1.2. pip versions <= 20.2 don't resolve dependencies conflicts correctly which could result in an incompatible numpy version and the following error:

  NotImplementedError: Cannot convert a symbolic Tensor (strided_slice_6:0) to a numpy array. This error may indicate that you're trying to pass a Tensor to a NumPy call, which is not supported

[2.2.9] - 2021-02-02

Bugfixes

• Correctly include the confused_with field in the test report for the POST /model/test/intents endpoint.

[2.2.8] - 2021-01-28

Bugfixes

• Fixes a bug in forms where the next slot asked was not consistent after returning to a form from an unhappy path.

[2.2.7] - 2021-01-25

Improvements

• Add support for in RasaYAMLWriter for writing intent and example metadata back into NLU YAML files.

Bugfixes

• Fixed a bug with Domain.is_domain_file() that could raise an Exception in case the potential domain file is not a valid YAML.

[2.2.6] - 2021-01-21

Bugfixes

• Fix wrong warning The method 'EventBroker.close' was changed to be asynchronous when the EventBroker.close was actually asynchronous.

• Fix incremental training for cases when training data does not contain entities but DIETClassifier is configured to perform entity recognition also.

  Now, the instance of RasaModelData inside DIETClassifier does not contain entities as a feature for training if there is no training data present for entity recognition.

[2.2.5] - 2021-01-12

Bugfixes

• Fixed key-error bug on rasa data validate stories.

Miscellaneous internal changes

Miscellaneous internal changes.

[2.2.4] - 2021-01-08

Improvements

• Improve the warning in case the RulePolicy or the deprecated MappingPolicy are missing from the model's policies configuration. Changed the info log to a warning as one of this policies should be added to the model configuration.

Bugfixes

• Explicitly specify the crypto extra dependency of pyjwt to ensure that the cryptography dependency is installed. cryptography is strictly required to be able to be able to verify JWT tokens.

[2.2.3] - 2021-01-06

Bugfixes

• Correctly retrieve intent ranking from UserUttered even during default affirmation action implementation.

• Fixed a problem when using the POST /model/test/intents endpoint together with a model server. The error looked as follows:

  ERROR    rasa.core.agent:agent.py:327 Could not load model due to Detected inconsistent loop usage. Trying to schedule a task on a new event loop, but scheduler was created with a different event loop. Make sure there is only one event loop in use and that the scheduler is running on that one.

  This also fixes a problem where testing a model from a model server would change the production model.

[2.2.2] - 2020-12-21

Bugfixes

• Fixed incompatibility between Rasa Open Source 2.2.x and Rasa X < 0.35.

[2.2.1] - 2020-12-17

Bugfixes

• Fixed a problem where a form wouldn't reject when the FormValidationAction re-implemented required_slots.

• Fixed an error when using the SQLTrackerStore with a Postgres database and the parameter login_db specified.

  The error was:

  psycopg2.errors.SyntaxError: syntax error at end of input
  rasa-production_1  | LINE 1: SELECT 1 FROM pg_catalog.pg_database WHERE datname = ?

[2.2.0] - 2020-12-16

Deprecations and Removals

• Domain.random_template_for is deprecated and will be removed in Rasa Open Source 3.0.0. You can alternatively use the TemplatedNaturalLanguageGenerator.

  Domain.action_names is deprecated and will be removed in Rasa Open Source 3.0.0. Please use Domain.action_names_or_texts instead.

• Interfaces for Policy.__init__ and Policy.load have changed. See migration guide for details.

• Deprecate training and test data in Markdown format. This includes:

  • reading and writing of story files in Markdown format
  • reading and writing of NLU data in Markdown format
  • reading and writing of retrieval intent data in Markdown format

  Support for Markdown data will be removed entirely in Rasa Open Source 3.0.0.

  Please convert your existing Markdown data by using the commands from the migration guide:

  rasa data convert nlu -f yaml --data={SOURCE_DIR} --out={TARGET_DIR}
  rasa data convert nlg -f yaml --data={SOURCE_DIR} --out={TARGET_DIR}
  rasa data convert core -f yaml --data={SOURCE_DIR} --out={TARGET_DIR}

• Domain.add_categorical_slot_default_value, Domain.add_requested_slot and Domain.add_knowledge_base_slots are deprecated and will be removed in Rasa Open Source 3.0.0. Their internal versions are now called during the Domain creation. Calling them manually is no longer required.

Features

• Incremental training of models in a pipeline is now supported.

  If you have added new NLU training examples or new stories/rules for dialogue manager, you don't need to train the pipeline from scratch. Instead, you can initialize the pipeline with a previously trained model and continue finetuning the model on the complete dataset consisting of new training examples. To do so, use rasa train --finetune. For more detailed explanation of the command, check out the docs on incremental training.

  Added a configuration parameter additional_vocabulary_size to CountVectorsFeaturizer and number_additional_patterns to RegexFeaturizer. These parameters are useful to configure when using incremental training for your pipelines.

• Add the option to use cross-validation to the POST /model/test/intents endpoint. To use cross-validation specify the query parameter cross_validation_folds in addition to the training data in YAML format.

  Add option to run NLU evaluation (POST /model/test/intents) and model training (POST /model/train) asynchronously. To trigger asynchronous processing specify a callback URL in the query parameter callback_url which Rasa Open Source should send the results to. This URL will also be called in case of errors.

• Make TED Policy an end-to-end policy. Namely, make it possible to train TED on stories that contain intent and entities or user text and bot actions or bot text. If you don't have text in your stories, TED will behave the same way as before. Add possibility to predict entities using TED.

  Here's an example of a dialogue in the Rasa story format:

  stories:
  - story: collect restaurant booking info  # name of the story - just for debugging
    steps:
    - intent: greet                          # user message with no entities
    - action: utter_ask_howcanhelp           # action that the bot should execute
    - intent: inform                         # user message with entities
      entities:
      - location: "rome"
      - price: "cheap"
    - bot: On it                             # actual text that bot can output
    - action: utter_ask_cuisine
    - user: I would like [spanish](cuisine). # actual text that user input
    - action: utter_ask_num_people

  Some model options for TEDPolicy got renamed. Please update your configuration files using the following mapping:

  Old model option	New model option
  transformer_size	dictionary “transformer_size” with keys
  	“text”, “action_text”, “label_action_text”, “dialogue”
  number_of_transformer_layers	dictionary “number_of_transformer_layers” with keys
  	“text”, “action_text”, “label_action_text”, “dialogue”
  dense_dimension	dictionary “dense_dimension” with keys
  	“text”, “action_text”, “label_action_text”, “intent”,
  	“action_name”, “label_action_name”, “entities”, “slots”,
  	“active_loop”

Improvements

• Added a message showing the location where the failed stories file was saved.

• Add support for the top-level response keys quick_replies, attachment and elements refered to in rasa.core.channels.OutputChannel.send_reponse, as well as metadata.

• Changed the format of the histogram of confidence values for both correct and incorrect predictions produced by running rasa test.

• Run bandit checks on pull requests. Introduce make static-checks command to run all static checks locally.

• Add rasa train --dry-run command that allows to check if training needs to be performed and what exactly needs to be retrained.

• POST /model/test/intents now returns the report field for intent_evaluation, entity_evaluation and response_selection_evaluation as machine-readable JSON payload instead of string.

• Make rasa data validate stories work for end-to-end.

  The rasa data validate stories function now considers the tokenized user text instead of the plain text that is part of a state. This is closer to what Rasa Core actually uses to distinguish states and thus captures more story structure problems.

Bugfixes

• Rename language_list to supported_language_list for JiebaTokenizer.
• A float slot returns unambiguous values - [1.0, <value>] if successfully converted, [0.0, 0.0] if not. This makes it possible to distinguish an empty float slot from a slot set to 0.0.
  caution

  This change is model-breaking. Please retrain your models.

• Fix an erroneous attribute for Redis key prefix in rasa.core.tracker_store.RedisTrackerStore: 'RedisTrackerStore' object has no attribute 'prefix'.
• Remove token when its text (for example, whitespace) can't be tokenized by LM tokenizer (from LanguageModelFeaturizer).
• Temporary directories which were created during requests to the HTTP API are now cleaned up correctly once the request was processed.
• Add option use_word_boundaries for RegexFeaturizer and RegexEntityExtractor. To correctly process languages such as Chinese that don't use whitespace for word separation, the user needs to add the use_word_boundaries: False option to those two components.
• Correctly fingerprint the default domain slots. Previously this led to the issue that rasa train core would always retrain the model even if the training data hasn't changed.

Improved Documentation

• Return the "Migrate from" entry to the docs sidebar.

Miscellaneous internal changes

Miscellaneous internal changes.

[2.1.3] - 2020-12-04

Improvements

• Removed multidict from the project dependencies. multidict continues to be a second order dependency of Rasa Open Source but will be determined by the dependencies which use it instead of by Rasa Open Source directly.

  This resolves issues like the following:

  sanic 20.9.1 has requirement multidict==5.0.0, but you'll have multidict 4.6.0 which is incompatible.

Bugfixes

• SingleStateFeaturizer checks whether it was trained with RegexInterpreter as nlu interpreter. If that is the case, RegexInterpreter is used during prediction.

• Make sure the responses are synced between NLU training data and the Domain even if there're no retrieval intents in the NLU training data.

• Categorical slots will have a default value set when just updating nlg data in the domain.

  Previously this resulted in InvalidDomain being thrown.

• • Preserve domain slot ordering while dumping it back to the file.
  • Preserve multiline text examples of responses defined in domain and NLU training data.

[2.1.2] - 2020-11-27

Bugfixes

• Slots that use initial_value won't cause rule contradiction errors when conversation_start: true is used. Previously, two rules that differed only in their use of conversation_start would be flagged as contradicting when a slot used initial_value.

  In checking for incomplete rules, an action will be required to have set only those slots that the same action has set in another rule. Previously, an action was expected to have set also slots which, despite being present after this action in another rule, were not actually set by this action.

• Fixed Rasa Open Source not being able to fetch models from certain URLs.

[2.1.1] - 2020-11-23

Bugfixes

• Sender ID is correctly set when copying the tracker and sending it to the action server (instead of sending the default value). This fixes a problem where the action server would only retrieve trackers with a sender_id default.

[2.1.0] - 2020-11-17

Deprecations and Removals

• The Policy interface was changed to return a PolicyPrediction object when predict_action_probabilities is called. Returning a list of probabilities directly is deprecated and support for this will be removed in Rasa Open Source 3.0.

  You can adapt your custom policy by wrapping your probabilities in a PolicyPrediction object:

  from rasa.core.policies.policy import Policy, PolicyPrediction
  # ... other imports
  
  def predict_action_probabilities(
          self,
          tracker: DialogueStateTracker,
          domain: Domain,
          interpreter: NaturalLanguageInterpreter,
          **kwargs: Any,
      ) -> PolicyPrediction:
      probabilities = ... # an action prediction of your policy
      return PolicyPrediction(probabilities, "policy_name", policy_priority=self.priority)

  The same change was applied to the PolicyEnsemble interface. Instead of returning a tuple of action probabilities and policy name, it is now returning a PolicyPrediction object. Support for the old PolicyEnsemble interface will be removed in Rasa Open Source 3.0.

  caution

  This change is model-breaking. Please retrain your models.

• The Pika Event Broker no longer supports the environment variables RABBITMQ_SSL_CA_FILE and RABBITMQ_SSL_KEY_PASSWORD. You can alternatively specify RABBITMQ_SSL_CA_FILE in the RabbitMQ connection URL as described in the RabbitMQ documentation.

  event_broker:
   type: pika
   url: "amqps://user:password@host?cacertfile=path_to_ca_cert&password=private_key_password"
   queues:
   - my_queue
  

  Support for RABBITMQ_SSL_KEY_PASSWORD was removed entirely.

  The method Event Broker.close was changed to be asynchronous. Support for synchronous implementations will be removed in Rasa Open Source 3.0.0. To adapt your implementation add the async keyword:

  from rasa.core.brokers.broker import EventBroker
  
  class MyEventBroker(EventBroker):
  
      async def close(self) -> None:
          # clean up event broker resources

Features

• Policies can now return obligatory and optional events as part of their prediction. Obligatory events are always applied to the current conversation tracker. Optional events are only applied to the conversation tracker in case the policy wins.

Improvements

• Changed Agent.load method to support pathlib paths.

• If you are using the feature Entity Roles and Groups, you should now also list the roles and groups in your domain file if you want roles and groups to influence your conversations. For example:

  entities:
    - city:
        roles:
          - from
          - to
    - name
    - topping:
        groups:
          - 1
          - 2
    - size:
        groups:
          - 1
          - 2

  Entity roles and groups can now influence dialogue predictions. For more information see the section Entity Roles and Groups influencing dialogue predictions.

• Predictions of the FallbackClassifier are ignored when evaluating the NLU model Note that the FallbackClassifier predictions still apply to test stories.

• Adapt the training data reader and emulator for wit.ai to their latest format. Update the instructions in the migrate from wit.ai documentation to run Rasa Open Source in wit.ai emulation mode.

• Adding configurable prefixes to Redis Tracker and Lock Stores so that a single Redis instance (and logical DB) can support multiple conversation trackers and locks. By default, conversations will be prefixed with tracker:... and all locks prefixed with lock:.... Additionally, you can add an alphanumeric-only prefix: value in endpoints.yml such that keys in redis will take the form value:tracker:... and value:lock:... respectively.

• Log the model's relative path when using CLI commands.

• Adds the option to configure whether extracted entities should be split by comma (",") or not. The default behaviour is True - i.e. split any list of extracted entities by comma. This makes sense for a list of ingredients in a recipie, for example "avocado, tofu, cauliflower", however doesn't make sense for an address such as "Schönhauser Allee 175, 10119 Berlin, Germany".

  In the latter case, add a new option to your config, e.g. if you are using the DIETClassifier this becomes:

  ...
  - name: DIETClassifier
    split_entities_by_comma: False
  ...

  in which case, none of the extracted entities will be split by comma. To switch it on/off for specific entity types you can use:

  ...
  - name: DIETClassifier
    split_entities_by_comma:
      address: True
      ingredient: False
  ...

  where both address and ingredient are two entity types.

  This feature is also available for CRFEntityExtractor.

• Fetching test stories from the HTTP API endpoint GET /conversations/<conversation_id>/story no longer triggers an update of the conversation session.

  Added a new boolean query parameter all_sessions (default: false) to the HTTP API endpoint for fetching test stories (GET /conversations/<conversation_id>/story).

  When setting ?all_sessions=true, the endpoint returns test stories for all conversation sessions for conversation_id. When setting ?all_sessions=all_sessions, or when omitting the all_sessions parameter, a single test story is returned for conversation_id. In cases where multiple conversation sessions exist, only the last story is returned.

  Specifying the retrieve_events_from_previous_conversation_sessions kwarg for the Tracker Store class is deprecated and will be removed in Rasa Open Source 3.0. Please use the retrieve_full_tracker() method instead.

• Improve the rasa data convert nlg command and introduce the rasa data convert responses command to simplify the migration from pre-2.0 response selector format to the new format.

• Added warning for when an option is provided for a component that is not listed as a key in the defaults for that component.

• Forms no longer reject their execution before a potential custom action for validating / extracting slots was executed. Forms continue to reject in two cases automatically:

  • A slot was requested to be filled, but no slot mapping applied to the latest user message and there was no custom action for potentially extracting other slots.
  • A slot was requested to be filled, but the custom action for validating / extracting slots didn't return any slot event.

  Additionally you can also reject the form execution manually by returning a ActionExecutionRejected event within your custom action for validating / extracting slots.

• Remove dependency between ConveRTTokenizer and ConveRTFeaturizer. The ConveRTTokenizer is now deprecated, and the ConveRTFeaturizer can be used with any other Tokenizer.

  Remove dependency between HFTransformersNLP, LanguageModelTokenizer, and LanguageModelFeaturizer. Both HFTransformersNLP and LanguageModelTokenizer are now deprecated. LanguageModelFeaturizer implements the behavior of the stack and can be used with any other Tokenizer.

• Gray out "Download" button in Rasa Playground when the project is not yet ready to be downloaded.

• Slot mappings for Forms in the domain are now optional. If you do not provide any slot mappings as part of the domain, you need to provide custom slot mappings through a custom action. A form without slot mappings is specified as follows:

  forms:
    my_form:
      # no mappings

  The action for forms can now be overridden by defining a custom action with the same name as the form. This can be used to keep using the deprecated Rasa Open Source FormAction which is implemented within the Rasa SDK. Note that it is not recommended to override the form action for anything else than using the deprecated Rasa SDK FormAction.

• Changed the default model weights loaded for HFTransformersNLP component.

  Use a language agnostic sentence embedding model as the default model. These model weights should help improve performance on intent classification and response selection.

• Add validations for slot mappings. If a slot mapping is not valid, an InvalidDomain error is raised.

• Adapt the training data reader and emulator for LUIS to their v3 format and add support for roles. Update the instructions in the "Migrate from LUIS" documentation page to reflect the recent changes made to the UI of LUIS.

• Adapt the training data reader and emulator for DialogFlow to their latest format and add support for regex entities.

• The Pika Event Broker was reimplemented with the [aio-pika library[(https://docs.aio-pika.com/). Messages will now be published to RabbitMQ asynchronously which improves the prediction performance.

• The confidence of the FallbackClassifier predictions is set to 1 - top intent confidence.

Bugfixes

• ActionRestart will now trigger ActionSessionStart as a followup action.

• Fixed a bug with rasa data split nlu which caused the resulting train / test ratio to sometimes differ from the ratio specified by the user or by default.

  The splitting algorithm ensures that every intent and response class appears in both the training and the test set. This means that each split must contain at least as many examples as there are classes, which for small datasets can contradict the requested training fraction. When this happens, the command issues a warning to the user that the requested training fraction can't be satisfied.

• Fixed bug where slots with influence_conversation=false affected the action prediction if they were set manually using the POST /conversations/<conversation_id/tracker/events endpoint in the HTTP API.

• Update Pika event broker to be a separate process and make it use a multiprocessing.Queue to send and process messages. This change should help avoid situations when events stop being sent after a while.

• Ignore rules when validating stories

• • Updated Slack Connector for new Slack Events API

• Update Rasa Playground "Download" button to work correctly depending on the current chat state.

• Test stories can now contain both: normal intents and retrieval intents. The failed_test_stories.yml, generated by rasa test, also specifies the full retrieval intent now. Previously rasa test would fail on test stories that specified retrieval intents.

• The converter tool is now able to convert test stories that contain a number as entity type.

• The converter tool now converts test stories and stories that contain full retrieval intents correctly. Previously the response keys were deleted during conversion to YAML.

• The slack connector requires a configuration for slack_signing_secret to make the connector more secure. The configuration value needs to be added to your credentials.yml if you are using the slack connector.

• Fixed model fingerprinting - it should avoid some more unecessary retrainings now.

• Fixed a problem when slots of type text or list were referenced by name only in the training data and this was treated as an empty value. This means that the two following stories are equivalent in case the slot type is text:

  stories:
  - story: Story referencing slot by name
    steps:
    - intent: greet
    - slot_was_set:
      - name
  
  - story: Story referencing slot with name and value
    steps:
    - intent: greet
    - slot_was_set:
      - name: "some name"
  

  Note that you still need to specify values for all other slot types as only text and list slots are featurized in a binary fashion.

Improved Documentation

• Correct data validation docs

Miscellaneous internal changes

Miscellaneous internal changes.

[2.0.8] - 2020-11-26

Bugfixes

• Slots that use initial_value won't cause rule contradiction errors when conversation_start: true is used. Previously, two rules that differed only in their use of conversation_start would be flagged as contradicting when a slot used initial_value.

  In checking for incomplete rules, an action will be required to have set only those slots that the same action has set in another rule. Previously, an action was expected to have set also slots which, despite being present after this action in another rule, were not actually set by this action.

[2.0.7] - 2020-11-24

Bugfixes

• ActionRestart will now trigger ActionSessionStart as a followup action.

• Fixed Rasa Open Source not being able to fetch models from certain URLs.

  This addresses an issue introduced in 2.0.3 where rasa-production could not use the models from rasa-x in Rasa X server mode.

• SingleStateFeaturizer checks whether it was trained with RegexInterpreter as NLU interpreter. If that is the case, RegexInterpreter is used during prediction.

[2.0.6] - 2020-11-10

Bugfixes

• Fixed a bug that occurred when setting multiple Sanic workers in combination with a custom Lock Store. Previously, if the number was set higher than 1 and you were using a custom lock store, it would reject because of a strict check to use a Redis Lock Store.
• Fixed a bug in the TwoStageFallback action which reverted too many events after the user successfully rephrased.

[2.0.5] - 2020-11-10

Bugfixes

• Fix a bug because of which only one retrieval intent was present in all_retrieval_intent key of the output of ResponseSelector even if there were multiple retrieval intents present in the training data.

[2.0.4] - 2020-11-08

Bugfixes

• Fixed error when starting Rasa X locally without a proper git setup.
• Properly validate incoming webhook requests for the Slack connector to be authentic.

[2.0.3] - 2020-10-29

Bugfixes

• Fix ConveRTTokenizer failing because of wrong model URL by making the model_url parameter of ConveRTTokenizer mandatory.

  Since the ConveRT model was taken offline, we can no longer use the earlier public URL of the model. Additionally, since the licence for the model is unknown, we cannot host it ourselves. Users can still use the component by setting model_url to a community/self-hosted model URL or path to a local directory containing model files. For example:

  pipeline:
      - name: ConveRTTokenizer
        model_url: <remote/local path to model>

• Update example formbot to use FormValidationAction for slot validation

[2.0.2] - 2020-10-22

Bugfixes

• Fix description of previous event in output of rasa data validate stories
• Fixed command line coloring for windows command lines running an encoding other than utf-8.

Miscellaneous internal changes

Miscellaneous internal changes.

[2.0.1] - 2020-10-20

Bugfixes

• Create correct KafkaProducer for PLAINTEXT and SASL_SSL security protocols.
• • Fix YAMLStoryReader not being able to represent OR statements in conversion mode.
  • Fix MarkdownStoryWriter not being able to write stories with OR statements (when loaded in conversion mode).

[2.0.0] - 2020-10-07

Deprecations and Removals

• Removed previously deprecated packages rasa_nlu and rasa_core.

  Use imports from rasa.core and rasa.nlu instead.

• Removed previously deprecated classes:

  • event brokers (EventChannel and FileProducer, KafkaProducer, PikaProducer, SQLProducer)
  • intent classifier EmbeddingIntentClassifier
  • policy KerasPolicy

  Removed previously deprecated methods:

  • Agent.handle_channels
  • TrackerStore.create_tracker_store

  Removed support for pipeline templates in config.yml

  Removed deprecated training data keys entity_examples and intent_examples from json training data format.

• Removed restaurantbot example as it was confusing and not a great way to build a bot.

• LabelTokenizerSingleStateFeaturizer is deprecated. To replicate LabelTokenizerSingleStateFeaturizer functionality, add a Tokenizer with intent_tokenization_flag: True and CountVectorsFeaturizer to the NLU pipeline. An example of elements to be added to the pipeline is shown in the improvement changelog 6296`.

  BinarySingleStateFeaturizer is deprecated and will be removed in the future. We recommend to switch to SingleStateFeaturizer.

• Specifying the parameters force and save_to_default_model_directory as part of the JSON payload when training a model using POST /model/train is now deprecated. Please use the query parameters force_training and save_to_default_model_directory instead. See the API documentation for more information.

• The conversation event form was renamed to active_loop. Rasa Open Source will continue to be able to read and process old form events. Note that serialized trackers will no longer have the active_form field. Instead the active_loop field will contain the same information. Story representations in Markdown and YAML will use active_loop instead of form to represent the event.

• Removed support for queue argument in PikaEventBroker (use queues instead).

  Domain file:

  • Removed support for templates key (use responses instead).
  • Removed support for string responses (use dictionaries instead).

  NLU Component:

  • Removed support for provides attribute, it's not needed anymore.
  • Removed support for requires attribute (use required_components() instead).

  Removed _guess_format() utils method from rasa.nlu.training_data.loading (use guess_format instead).

  Removed several config options for TED Policy, DIETClassifier and ResponseSelector:

  • hidden_layers_sizes_pre_dial
  • hidden_layers_sizes_bot
  • droprate
  • droprate_a
  • droprate_b
  • hidden_layers_sizes_a
  • hidden_layers_sizes_b
  • num_transformer_layers
  • num_heads
  • dense_dim
  • embed_dim
  • num_neg
  • mu_pos
  • mu_neg
  • use_max_sim_neg
  • C2
  • C_emb
  • evaluate_every_num_epochs
  • evaluate_on_num_examples

  Please check the documentation for more information.

• The conversation event form_validation was renamed to loop_interrupted. Rasa Open Source will continue to be able to read and process old form_validation events.

• SklearnPolicy was deprecated. TEDPolicy is the preferred machine-learning policy for dialogue models.

• Slots of type unfeaturized are now deprecated and will be removed in Rasa Open Source 3.0. Instead you should use the property influence_conversation: false for every slot type as described in the migration guide.

• Conversation sessions are now enabled by default if your Domain does not contain a session configuration. Previously a missing session configuration was treated as if conversation sessions were disabled. You can explicitly disable conversation sessions using the following snippet:

  domain.yml

  session_config:
    # A session expiration time of `0`
    # disables conversation sessions
    session_expiration_time: 0

• Using the default action action_deactivate_form to deactivate the currently active loop / Form is deprecated. Please use action_deactivate_loop instead.

Features

• Added template name to the metadata of bot utterance events.

  BotUttered event contains a template_name property in its metadata for any new bot message.

• Added a --num-threads CLI argument that can be passed to rasa train and will be used to train NLU components.

• You can now define what kind of features should be used by what component (see Choosing a Pipeline).

  You can set an alias via the option alias for every featurizer in your pipeline. The alias can be anything, by default it is set to the full featurizer class name. You can then specify, for example, on the DIETClassifier what features from which featurizers should be used. If you don't set the option featurizers all available features will be used. This is also the default behavior. Check components to see what components have the option featurizers available.

  Here is an example pipeline that shows the new option. We define an alias for all featurizers in the pipeline. All features will be used in the DIETClassifier. However, the ResponseSelector only takes the features from the ConveRTFeaturizer and the CountVectorsFeaturizer (word level).

  pipeline:
  - name: ConveRTTokenizer
  - name: ConveRTFeaturizer
    alias: "convert"
  - name: CountVectorsFeaturizer
    alias: "cvf_word"
  - name: CountVectorsFeaturizer
    alias: "cvf_char"
    analyzer: char_wb
    min_ngram: 1
    max_ngram: 4
  - name: RegexFeaturizer
    alias: "regex"
  - name: LexicalSyntacticFeaturizer
    alias: "lsf"
  - name: DIETClassifier:
  - name: ResponseSelector
    epochs: 50
    featurizers: ["convert", "cvf_word"]
  - name: EntitySynonymMapper
  caution

  This change is model-breaking. Please retrain your models.

• Added --port commandline argument to the interactive learning mode to allow changing the port for the Rasa server running in the background.

• Add new entity extractor RegexEntityExtractor. The entity extractor extracts entities using the lookup tables and regexes defined in the training data. For more information see RegexEntityExtractor.

• Introduced a new YAML format for Core training data and implemented a parser for it. Rasa Open Source can now read stories in both Markdown and YAML format.

• You can now enable threaded message responses from Rasa through the Slack connector. This option is enabled using an optional configuration in the credentials.yml file

      slack:
        slack_token:
        slack_channel:
        use_threads: True

  Button support has also been added in the Slack connector.

• Add support for rules data and forms in YAML format.

• The NLU interpreter is now passed to the Policies during training and inference time. Note that this requires an additional parameter interpreter in the method predict_action_probabilities of the Policy interface. In case a custom Policy implementation doesn't provide this parameter Rasa Open Source will print a warning and omit passing the interpreter.

• Added the new dialogue policy RulePolicy which will replace the old “rule-like” policies Mapping Policy, Fallback Policy, Two-Stage Fallback Policy, and Form Policy. These policies are now deprecated and will be removed in the future. Please see the rules documentation for more information.

  Added new NLU component FallbackClassifier which predicts an intent nlu_fallback in case the confidence was below a given threshold. The intent nlu_fallback may then be used to write stories / rules to handle the fallback in case of low NLU confidence.

  pipeline:
  - # Other NLU components ...
  - name: FallbackClassifier
    # If the highest ranked intent has a confidence lower than the threshold then
    # the NLU pipeline predicts an intent `nlu_fallback` which you can then be used in
    # stories / rules to implement an appropriate fallback.
    threshold: 0.5

• Added possibility to split the domain into separate files. All YAML files under the path specified with --domain will be scanned for domain information (e.g. intents, actions, etc) and then combined into a single domain.

  The default value for --domain is still domain.yml.

• Add optional metadata argument to NaturalLanguageInterpreter's parse method.

• The Rasa Open Source API endpoint POST /model/train now supports training data in YAML format. Please specify the header Content-Type: application/yaml when training a model using YAML training data. See the API documentation for more information.

• Added a YAML schema and a writer for 2.0 Training Core data.

• Users can now use the rasa data convert \{nlu|core\} -f yaml command to convert training data from Markdown format to YAML format.

• Add option use_lemma to CountVectorsFeaturizer. By default it is set to True.

  use_lemma indicates whether the featurizer should use the lemma of a word for counting (if available) or not. If this option is set to False it will use the word as it is.

Improvements

• Add support for Python 3.8.

• Changed the project structure for Rasa projects initialized with the CLI (using the rasa init command): actions.py -> actions/actions.py. actions is now a Python package (it contains a file actions/__init__.py). In addition, the __init__.py at the root of the project has been removed.

• DIETClassifier now also assigns a confidence value to entity predictions.

• Added behavior to the rasa --version command. It will now also list information about the operating system, python version and rasa-sdk. This will make it easier for users to file bug reports.

• Support for additional training metadata.

  Training data messages now to support kwargs and the Rasa JSON data reader includes all fields when instantiating a training data instance.

• Standardize testing output. The following test output can be produced for intents, responses, entities and stories:

  • report: a detailed report with testing metrics per label (e.g. precision, recall, accuracy, etc.)
  • errors: a file that contains incorrect predictions
  • successes: a file that contains correct predictions
  • confusion matrix: plot of confusion matrix
  • histogram: plot of confidence distribution (not available for stories)

• To avoid the problem of our entity extractors predicting entity labels for just a part of the words, we introduced a cleaning method after the prediction was done. We should avoid the incorrect prediction in the first place. To achieve this we will not tokenize words into sub-words anymore. We take the mean feature vectors of the sub-words as the feature vector of the word.

  caution

  This change is model breaking. Please, retrain your models.

• Move option case_sensitive from the tokenizers to the featurizers.

  • Remove the option from the WhitespaceTokenizer and ConveRTTokenizer.
  • Add option case_sensitive to the RegexFeaturizer.

• If a user sends a voice message to the bot using Facebook, users messages was set to the attachments URL. The same is now also done for the rest of attachment types (image, video, and file).

• Creating a Domain using Domain.fromDict can no longer alter the input dictionary. Previously, there could be problems when the input dictionary was re-used for other things after creating the Domain from it.

• The debug-level logs when instantiating an SQLTrackerStore no longer show the password in plain text. Now, the URL is displayed with the password hidden, e.g. postgresql://username:***@localhost:5432.

• Shorten the information in tqdm during training ML algorithms based on the log level. If you train your model in debug mode, all available metrics will be shown during training, otherwise, the information is shorten.

• Ignore conversation test directory tests/ when importing a project using MultiProjectImporter and use_e2e is False. Previously, any story data found in a project subdirectory would be imported as training data.

• Implemented model checkpointing for DIET (including the response selector) and TED. The best model during training will be stored instead of just the last model. The model is evaluated on the basis of evaluate_every_number_of_epochs and evaluate_on_number_of_examples.

  Checkpointing is enabled iff the following is set for the models in the config.yml file:

  • checkpoint_model: True
  • evaluate_on_number_of_examples > 0

  The model is stored to whatever location has been specified with the --out parameter when calling rasa train nlu/core ....

• rasa data split nlu now makes sure that there is at least one example per intent and response in the test data.

• The method ensure_consistent_bilou_tagging now also considers the confidence values of the predicted tags when updating the BILOU tags.

• We updated the way how we save and use features in our NLU pipeline.

  The message object now has a dedicated field, called features, to store the features that are generated in the NLU pipeline. We adapted all our featurizers in a way that sequence and sentence features are stored independently. This allows us to keep different kind of features for the sequence and the sentence. For example, the LexicalSyntacticFeaturizer does not produce any sentence features anymore as our experiments showed that those did not bring any performance gain just quite a lot of additional values to store.

  We also modified the DIET architecture to process the sequence and sentence features independently at first. The features are concatenated just before the transformer.

  We also removed the __CLS__ token again. Our Tokenizers will not add this token anymore.

  caution

  This change is model-breaking. Please retrain your models.

• Add endpoint kwarg to rasa.jupyter.chat to enable using a custom action server while chatting with a model in a jupyter notebook.

• Support for rasa conversation id with special characters on the server side - necessary for some channels (e.g. Viber)

• Add support for proxy use in slack input channel.

• Log the number of examples per intent during training. Logging can be enabled using rasa train --debug.

• Support for other remote storages can be achieved by using an external library.

• Add output_channel query param to /conversations/<conversation_id>/tracker/events route, along with boolean execute_side_effects to optionally schedule/cancel reminders, and forward bot messages to output channel.

• Allow Rasa to boot when model loading exception occurs. Forward HTTP Error responses to standard log output.

• Rename DucklingHTTPExtractor to DucklingEntityExtractor.

• • Modified functionality of SingleStateFeaturizer.

    SingleStateFeaturizer uses trained NLU Interpreter to featurize intents and action names. This modified SingleStateFeaturizer can replicate LabelTokenizerSingleStateFeaturizer functionality. This component is deprecated from now on. To replicate LabelTokenizerSingleStateFeaturizer functionality, add a Tokenizer with intent_tokenization_flag: True and CountVectorsFeaturizer to the NLU pipeline. Please update your configuration file.

    For example:

    language: en
    pipeline:
      - name: WhitespaceTokenizer
        intent_tokenization_flag: True
      - name: CountVectorsFeaturizer

    Please train both NLU and Core (using rasa train) to use a trained tokenizer and featurizer for core featurization.

    The new SingleStateFeaturizer stores slots, entities and forms in sparse features for more lightweight storage.

    BinarySingleStateFeaturizer is deprecated and will be removed in the future. We recommend to switch to SingleStateFeaturizer.

  • Modified TEDPolicy to handle sparse features. As a result, TEDPolicy may require more epochs than before to converge.

  • Default TEDPolicy featurizer changed to MaxHistoryTrackerFeaturizer with infinite max history (takes all dialogue turns into account).

  • Default batch size for TED increased from [8,32] to [64, 256]

• Response selector templates now support all features that domain utterances do. They use the yaml format instead of markdown now. This means you can now use buttons, images, ... in your FAQ or chitchat responses (assuming they are using the response selector).

  As a consequence, training data form in markdown has to have the file suffix .md from now on to allow proper file type detection-

• Support for test stories written in yaml format.

• Response Selectors are now trained on retrieval intent labels by default instead of the actual response text. For most models, this should improve training time and accuracy of the ResponseSelector.

  If you want to revert to the pre-2.0 default behavior, add the use_text_as_label=true parameter to your ResponseSelector component.

  You can now also have multiple response templates for a single sub-intent of a retrieval intent. The first response template containing the text attribute is picked for training(if use_text_as_label=True) and a random template is picked for bot's utterance just as how other utter_ templates are picked.

  All response selector related evaluation artifacts - report.json, successes.json, errors.json, confusion_matrix.png now use the sub-intent of the retrieval intent as the target and predicted labels instead of the actual response text.

  The output schema of ResponseSelector has changed - full_retrieval_intent and name have been deprecated in favour of intent_response_key and response_templates respectively. Additionally a key all_retrieval_intents is added to the response selector output which will hold a list of all retrieval intents(faq,chitchat, etc.) that are present in the training data.An example output looks like this -

  "response_selector": {
      "all_retrieval_intents": ["faq"],
      "default": {
        "response": {
          "id": 1388783286124361986, "confidence": 1.0, "intent_response_key": "faq/is_legit",
          "response_templates": [
            {
              "text": "absolutely",
              "image": "https://i.imgur.com/nGF1K8f.jpg"
            },
            {
              "text": "I think so."
            }
          ],
        },
        "ranking": [
          {
            "id": 1388783286124361986,
            "confidence": 1.0,
            "intent_response_key": "faq/is_legit"
          },
        ]

  An example bot demonstrating how to use the ResponseSelector is added to the examples folder.

• Do not modify conversation tracker's latest_input_channel property when using POST /trigger_intent or ReminderScheduled.

• Do not set the output dimension of the sparse-to-dense layers to the same dimension as the dense features.

  Update default value of dense_dimension and concat_dimension for text in DIETClassifier to 128.

• Retrieval actions with respond_ prefix are now replaced with usual utterance actions with utter_ prefix.

  If you were using retrieval actions before, rename all of them to start with utter_ prefix. For example, respond_chitchat becomes utter_chitchat. Also, in order to keep the response templates more consistent, you should now add the utter_ prefix to all response templates defined for retrieval intents. For example, a response template chitchat/ask_name becomes utter_chitchat/ask_name. Note that the NLU examples for this will still be under chitchat/ask_name intent. The example responseselectorbot should help clarify these changes further.

• Added telemetry reporting. Rasa uses telemetry to report anonymous usage information. This information is essential to help improve Rasa Open Source for all users. Reporting will be opt-out. More information can be found in our telemetry documentation.

• Update extract_other_slots method inside FormAction to fill a slot from an entity with a different name if corresponding slot mapping of from_entity type is unique.

• Slots of any type can now be ignored during a conversation. To do so, specify the property influence_conversation: false for the slot.

  slot:
    a_slot:
      type: text
      influence_conversation: false

  The property influence_conversation is set to true by default. See the documentation for slots for more information.

  A new slot type any was added. Slots of this type can store any value. Slots of type any are always ignored during conversations.

• Improved exception handling within Rasa Open Source.

  All exceptions that are somewhat expected (e.g. errors in file formats like configurations or training data) will share a common base class RasaException.

  ::warning Backwards Incompatibility Base class for the exception raised when an action can not be found has been changed from a NameError to a ValueError. ::

  Some other exceptions have also slightly changed:

  • raise YamlSyntaxException instead of YAMLError (from ruamel) when failing to load a yaml file with information about the line where loading failed
  • introduced MissingDependencyException as an exception raised if packages need to be installed

• Debug logs from matplotlib libraries are now hidden by default and are configurable with the LOG_LEVEL_LIBRARIES environment variable.

• Update KafkaEventBroker to support SASL_SSL and PLAINTEXT protocols.

Bugfixes

• Fixed issue where temporary model directories were not removed after pulling from a model server.

  If the model pulled from the server was invalid, this could lead to large amounts of local storage usage.

• Fixed a bug in the CountVectorsFeaturizer which resulted in the very first message after loading a model to be processed incorrectly due to the vocabulary not being loaded yet.

• Fixed Rasa shell skipping button messages if buttons are attached to a message previous to the latest.

• Stack level for FutureWarning updated to level 2.

• If custom utter message contains no value or integer value, then it fails returning custom utter message. Fixed by converting the template to type string.

• Don't create TensorBoard log files during prediction.

• Fixed DIET breaking with empty spaCy model.

• Pinned the library version for the Azure Cloud Storage to 2.1.0 since the persistor is currently not compatible with later versions of the azure-storage-blob library.

• Remove clean_up_entities from extractors that extract pre-defined entities. Just keep the clean up method for entity extractors that extract custom entities.

• Fixed issue where the DucklingHTTPExtractor component would not work if its url contained a trailing slash.

• Changed to variable CERT_URI in hangouts.py to a string type

• Slots will be correctly interpolated for button responses.

  Previously this resulted in no interpolation due to a bug.

• Remove option token_pattern from CountVectorsFeaturizer. Instead all tokenizers now have the option token_pattern. If a regular expression is set, the tokenizer will apply the token pattern.

• Allow user to retry failed file exports in interactive training.

• Fixed a bug when custom metadata passed with the utterance always restarted the session.

• WhitespaceTokenizer does not remove vowel signs in Hindi anymore.

• Convert entity values coming from DucklingHTTPExtractor to string during evaluation to avoid mismatches due to different types.

• Update FeatureSignature to store just the feature dimension instead of the complete shape. This change fixes the usage of the option share_hidden_layers in the DIETClassifier.

• Unescape the \n, \t, \r, \f, \b tokens on reading nlu data from markdown files.

  On converting json files into markdown, the tokens mentioned above are espaced. These tokens need to be unescaped on loading the data from markdown to ensure that the data is treated in the same way.

• Fix the way training data is generated in rasa test nlu when using the -P flag. Each percentage of the training dataset used to be formed as a part of the last sampled training dataset and not as a sample from the original training dataset.

• Prevent WhitespaceTokenizer from outputting empty list of tokens.

• Add EntityExtractor as a required component for EntitySynonymMapper in a pipeline.

• Better handling of input sequences longer than the maximum sequence length that the HFTransformersNLP models can handle.

  During training, messages with longer sequence length should result in an error, whereas during inference they are gracefully handled but a debug message is logged. Ideally, passing messages longer than the acceptable maximum sequence lengths of each model should be avoided.

• When using the DynamoTrackerStore, if there are more than 100 DynamoDB tables, the tracker could attempt to re-create an existing table if that table was not among the first 100 listed by the dynamo API.

• Fixed a deprication warning that pops up due to changes in numpy

• Update rasabaster to fix an issue with syntax highlighting on "Prototype an Assistant" page.

  Update default stories and rules on "Prototype an Assistant" page.

• Fixed a bug in the serialise method of the EvaluationStore class which resulted in a wrong end-to-end evaluation of the predicted entities.

• Forms with slot mappings defined in domain.yml must now be a dictionary (with form names as keys). The previous syntax where forms was simply a list of form names is still supported.

• Remove BILOU tag prefix from role and group labels when creating entities.

• Fixed a bug in the featurization of the boolean slot type. Previously, to set a slot value to "true", you had to set it to "1", which is in conflict with the documentation. In older versions true (without quotes) was also possible, but now raised an error during yaml validation.

• Fixed a bug in rasa interactive. Now it exports the stories and nlu training data as yml file.

• Fixed slots not being featurized before first user utterance.

  Fixed AugmentedMemoizationPolicy to forget the first action on the first going back

• Fixed the remote URL of ConveRT model as it was recently updated by its authors.

• Treat the length of OOV token as 1 to fix token align issue when OOV occurred.

• Fixed the bug when entity was extracted even if it had a role or group but roles or groups were not expected.

• Fixed the bug that caused supported_language_list of Component to not work correctly.

  To avoid confusion, only one of supported_language_list and not_supported_language_list can be set to not None now

• Fixed issue where responses including text: "" and no custom key would incorrectly fail domain validation.

• Fixed issue where extra keys other than title and payload inside of buttons made a response fail domain validation.

• Do not filter training data in model.py but on component side.

• Check if a model was provided when executing rasa test core. If not, print a useful error message and stop.

• Transfer only response templates for retrieval intents from domain to NLU Training Data.

  This avoids retraining the NLU model if one of the non retrieval intent response templates are edited.

Improved Documentation

• Added documentation on ambiguity_threshold parameter in Fallback Actions page.
• Remove outdated whitespace tokenizer warning in Testing Your Assistant documentation.
• Updated Facebook Messenger channel docs with supported attachment information
• Update rasa shell documentation to explain how to recreate external channel session behavior.
• Event brokers documentation should say url instead of host.
• Update rasa init documentation to include tests/conversation_tests.md in the resulting directory tree.
• Update "Validating Form Input" section to include details about how FormValidationAction class makes it easier to validate form slots in custom actions and how to use it.
• Update the examples in the API docs to use YAML instead of Markdown

Miscellaneous internal changes

Miscellaneous internal changes.

[1.10.26] - 2021-06-17

Features

• Added sasl_mechanism as an optional configurable parameter for the Kafka Producer.

[1.10.25] - 2021-04-14

Features

• Added partition_by_sender flag to Kafka Producer to optionally associate events with Kafka partition based on sender_id.

Improvements

• Improved the lock store debug log message when the process has to queue because other messages have to be processed before this item.

[1.10.24] - 2021-03-29

Bugfixes

• Added group_id parameter back to KafkaEventBroker to fix error when instantiating event broker with a config containing the group_id parameter which is only relevant to the event consumer

[1.10.23] - 2021-02-22

Bugfixes

• Fixed bug where the conversation does not lock before handling a reminder event.

[1.10.22] - 2021-02-05

Bugfixes

• Backported the Rasa Open Source 2 PikaEventBroker implementation to address problems when using it with multiple Sanic workers.

[1.10.21] - 2021-02-01

Improvements

• The url option now supports a list of servers url: ['10.0.0.158:32803','10.0.0.158:32804']. Removed group_id because it is not a valid Kafka producer parameter.

Bugfixes

• Fixed a bug that occurred when setting multiple Sanic workers in combination with a custom Lock Store. Previously, if the number was set higher than 1 and you were using a custom lock store, it would reject because of a strict check to use a Redis Lock Store.
• Fix a bug where, if a user injects an intent using the HTTP API, slot auto-filling is not performed on the entities provided.

[1.10.20] - 2020-12-18

Bugfixes

• Fix scikit-learn crashing during evaluation of ResponseSelector predictions.

[1.10.19] - 2020-12-17

Improvements

• Kafka Producer connection now remains active across sends. Added support for group and client id. The Kafka producer also adds support for the PLAINTEXT and SASL_SSL protocols.

  DynamoDB table exists check fixed bug when more than 100 tables exist.

• Replace use of python-telegram-bot package with pyTelegramBotAPI

• Use response selector keys (sub-intents) as labels for plotting the confusion matrix during NLU evaluation to improve readability.

[1.10.18] - 2020-11-26

Bugfixes

• #7340: Fixed an issues with the DynamoDB TrackerStore creating a new table entry/object for each TrackerStore update. The column session_date has been deprecated and should be removed manually in existing DynamoDB tables.

[1.10.17] - 2020-11-12

Bugfixes

• Prevent the message handling process in PikaEventBroker from being terminated.

[1.10.16] - 2020-10-15

Bugfixes

• Update Pika event broker to be a separate process and make it use a multiprocessing.Queue to send and process messages. This change should help avoid situations when events stop being sent after a while.

[1.10.15] - 2020-10-09

Bugfixes

• Fixed issue where temporary model directories were not removed after pulling from a model server. If the model pulled from the server was invalid, this could lead to large amounts of local storage usage.
• Treat the length of OOV token as 1 to fix token align issue when OOV occurred.
• Fixed MappingPolicy not predicting action_listen after the mapped action while running rasa test.

Improvements

• Debug logs from matplotlib libraries are now hidden by default and are configurable with the LOG_LEVEL_LIBRARIES environment variable.

[1.10.14] - 2020-09-23

Bugfixes

• Fixed the remote URL of ConveRT model as it was recently updated by its authors. Also made the remote URL configurable at runtime in the corresponding tokenizer's and featurizer's configuration.

[1.10.13] - 2020-09-22

Bugfixes

• Remove BILOU tag prefix from role and group labels when creating entities.

[1.10.12] - 2020-09-03

Bugfixes

• Fix slow training of CRFEntityExtractor when using Entity Roles and Groups.

[1.10.11] - 2020-08-21

Improvements

• Do not deepcopy slots when instantiating trackers. This leads to a significant speedup when training on domains with a large number of slots.

• Added more debugging logs to the Lock Stores to simplify debugging in case of

  connection problems.

  Added a new parameter socket_timeout to the RedisLockStore. If Redis doesn't answer within socket_timeout seconds to requests from Rasa Open Source, an error is raised. This avoids seemingly infinitely blocking connections and exposes connection problems early.

Bugfixes

• Fixed a bug where domain fields such as store_entities_as_slots were overridden with defaults and therefore ignored.
• If two entities are separated by a comma (or any other symbol), extract them as two separate entities.
• If two entities are separated by a single space and uses BILOU tagging, extract them as two separate entities based on their BILOU tags.

[1.10.10] - 2020-08-04

Bugfixes

• Fixed TypeError: expected string or bytes-like object issue caused by integer, boolean, and null values in templates.

[1.10.9] - 2020-07-29

Improvements

• Rasa Open Source will no longer add responses to the actions section of the domain when persisting the domain as a file. This addresses related problems in Rasa X when Integrated Version Control introduced big diffs due to the added utterances in the actions section.

Bugfixes

• Consider entity roles/groups during interactive learning.

[1.10.8] - 2020-07-15

Bugfixes

• Add 'Access-Control-Expose-Headers' for 'filename' header
• Fixed a bug where an invalid language variable prevents rasa from finding training examples when importing Dialogflow data.

[1.10.7] - 2020-07-07

Features

• Add not_supported_language_list to component to be able to define languages that a component can NOT handle.

  WhitespaceTokenizer is not able to process languages which are not separated by whitespace. WhitespaceTokenizer will throw an error if it is used with Chinese, Japanese, and Thai.

Bugfixes

• WhitespaceTokenizer only removes emoji if complete token matches emoji regex.

[1.10.6] - 2020-07-06

Bugfixes

• Prevent WhitespaceTokenizer from outputting empty list of tokens.

[1.10.5] - 2020-07-02

Bugfixes

• Explicitly remove all emojis which appear as unicode characters from the output of regex.sub inside WhitespaceTokenizer.

[1.10.4] - 2020-07-01

Bugfixes

• WhitespaceTokenizer does not remove vowel signs in Hindi anymore.

• Previously, specifying a lock store in the endpoint configuration with a type other than redis or in_memory would lead to an AttributeError: 'str' object has no attribute 'type'. This bug is fixed now.

• Fix Interpreter parsed an intent ... warning when using the /model/parse endpoint with an NLU-only model.

• Convert entity values coming from any entity extractor to string during evaluation to avoid mismatches due to different types.

• The assistant will respond through the webex channel to any user (room) communicating to it. Before the bot responded only to a fixed roomId set in the credentials.yml config file.

[1.10.3] - 2020-06-12

Improvements

• Reduced duplicate logs and warnings when running rasa train.

Bugfixes

• Remove the clean_up_entities method from the DIETClassifier and CRFEntityExtractor as it let to incorrect entity predictions.

• Fix server crashes that occurred when Rasa Open Source pulls a model from a model server and an exception was thrown during model loading (such as a domain with invalid YAML).

[1.10.2] - 2020-06-03

Bugfixes

• Responses used in ResponseSelector now support new lines with explicitly adding \\n between them.

• Fixed a bug in rasa export) which caused Rasa Open Source to only migrate conversation events from the last Session configuration.

[1.10.1] - 2020-05-15

Improvements

• Creating a Domain using Domain.fromDict can no longer alter the input dictionary. Previously, there could be problems when the input dictionary was re-used for other things after creating the Domain from it.

Bugfixes

• Don't create TensorBoard log files during prediction.

• Fix: DIET breaks with empty spaCy model

• Remove clean_up_entities from extractors that extract pre-defined entities. Just keep the clean up method for entity extractors that extract custom entities.

• Fixed issue where the DucklingHTTPExtractor component would not work if its url contained a trailing slash.

• Fix list index out of range error in ensure_consistent_bilou_tagging.

Miscellaneous internal changes

Miscellaneous internal changes.

[1.10.0] - 2020-04-28

Features

• Add support for entities with roles and grouping of entities in Rasa NLU.

  You can now define a role and/or group label in addition to the entity type for entities. Use the role label if an entity can play different roles in your assistant. For example, a city can be a destination or a departure city. The group label can be used to group multiple entities together. For example, you could group different pizza orders, so that you know what toppings goes with which pizza and what size which pizza has. For more details see Entities Roles and Groups.

  To fill slots from entities with a specific role/group, you need to either use forms or use a custom action. We updated the tracker method get_latest_entity_values to take an optional role/group label. If you want to use a form, you can add the specific role/group label of interest to the slot mapping function from_entity (see Forms).

  note

  Composite entities are currently just supported by the DIETClassifier and CRFEntityExtractor.

• Update training data format for NLU to support entities with a role or group label.

  You can now specify synonyms, roles, and groups of entities using the following data format: Markdown:

  [LA]{"entity": "location", "role": "city", "group": "CA", "value": "Los Angeles"}

  JSON:

  "entities": [
      {
          "start": 10,
          "end": 12,
          "value": "Los Angeles",
          "entity": "location",
          "role": "city",
          "group": "CA",
      }
  ]

  The markdown format [LA](location:Los Angeles) is deprecated. To update your training data file just execute the following command on the terminal of your choice: sed -i -E 's/\\[([^)]+)\\]\\(([^)]+):([^)]+)\\)/[\\1]{"entity": "\\2", "value": "\\3"}/g' nlu.md

  For more information about the new data format see Training Data Format.

Improvements

• Suppressed pika logs when establishing the connection. These log messages mostly happened when Rasa X and RabbitMQ were started at the same time. Since RabbitMQ can take a few seconds to initialize, Rasa X has to re-try until the connection is established. In case you suspect a different problem (such as failing authentication) you can re-enable the pika logs by setting the log level to DEBUG. To run Rasa Open Source in debug mode, use the --debug flag. To run Rasa X in debug mode, set the environment variable DEBUG_MODE to true.

• Include the source filename of a story in the failed stories

  Include the source filename of a story in the failed stories to make it easier to identify the file which contains the failed story.

• Add confusion matrix and “confused_with” to response selection evaluation

  If you are using ResponseSelectors, they now produce similiar outputs during NLU evaluation. Misclassfied responses are listed in a “confused_with” attribute in the evaluation report. Similiarily, a confusion matrix of all responses is plotted.

• Added socketio to the compatible channels for Reminders and External Events.

• Update POST /model/train endpoint to accept retrieval action responses at the responses key of the JSON payload.

• All Rasa Open Source images are now using Python 3.7 instead of Python 3.6.

• Update dependencies based on the dependabot check.

• Add dropout between FFNN and DenseForSparse layers in DIETClassifier, ResponseSelector and EmbeddingIntentClassifier controlled by use_dense_input_dropout config parameter.

• DIETClassifier only counts as extractor in rasa test if it was actually trained for entity recognition.

• Remove regularization gradient for variables that don't have prediction gradient.

• Raise a warning in CRFEntityExtractor and DIETClassifier if entities are not correctly annotated in the training data, e.g. their start and end values do not match any start and end values of tokens.

• Add full_retrieval_intent property to ResponseSelector rankings

• Change default values for hyper-parameters in EmbeddingIntentClassifier and DIETClassifier

  Use scale_loss=False in DIETClassifier. Reduce the number of dense dimensions for sparse features of text from 512 to 256 in EmbeddingIntentClassifier.

Bugfixes

• Fixed issue where posting to certain callback channel URLs would return a 500 error on successful posts due to invalid response format.

• One word can just have one entity label.

  If you are using, for example, ConveRTTokenizer words can be split into multiple tokens. Our entity extractors assign entity labels per token. So, it might happen, that a word, that was split into two tokens, got assigned two different entity labels. This is now fixed. One word can just have one entity label at a time.

• An entity label should always cover a complete word.

  If you are using, for example, ConveRTTokenizer words can be split into multiple tokens. Our entity extractors assign entity labels per token. So, it might happen, that just a part of a word has an entity label. This is now fixed. An entity label always covers a complete word.

• Fixed an issue that happened when metadata is passed in a new session.

  Now the metadata is correctly passed to the ActionSessionStart.

• Updated Python dependency ruamel.yaml to >=0.16. We recommend to use at least 0.16.10 due to the security issue CVE-2019-20478 which is present in in prior versions.

Miscellaneous internal changes

Miscellaneous internal changes.

[1.9.7] - 2020-04-23

Improvements

• The stream reading timeout for rasa shell\ is now configurable by using the environment variable ``RASA_SHELL_STREAM_READING_TIMEOUT_IN_SECONDS. This can help to fix problems when using rasa shell` with custom actions which run 10 seconds or longer.

Bugfixes

• Reverted changes in 1.9.6 that led to model incompatibility. Upgrade to 1.9.7 to fix self.sequence_lengths_for(tf_batch_data[TEXT_SEQ_LENGTH][0]) IndexError: list index out of range error without needing to retrain earlier 1.9 models.

  Therefore, all 1.9 models except for 1.9.6 will be compatible; a model trained on 1.9.6 will need to be retrained on 1.9.7.

[1.9.6] - 2020-04-15

Bugfixes

• Fix rasa test nlu plotting when using multiple runs.

• Fixed issue where max_number_of_predictions was not considered when running end-to-end testing.

Miscellaneous internal changes

Miscellaneous internal changes.

[1.9.5] - 2020-04-01

Improvements

• Support for PostgreSQL schemas in SQLTrackerStore. The SQLTrackerStore accesses schemas defined by the POSTGRESQL_SCHEMA environment variable if connected to a PostgreSQL database.

  The schema is added to the connection string option's -csearch_path key, e.g. -options=-csearch_path=<SCHEMA_NAME> (see the PostgreSQL docs for more details). As before, if no POSTGRESQL_SCHEMA is defined, Rasa uses the database's default schema (public).

  The schema has to exist in the database before connecting, i.e. it needs to have been created with

  CREATE SCHEMA schema_name;

Bugfixes

• Fixed ambiguous logging in DIETClassifier by adding the name of the calling class to the log message.

[1.9.4] - 2020-03-30

Bugfixes

• Fix memory leak problem on increasing number of calls to /model/parse endpoint.

[1.9.3] - 2020-03-27

Bugfixes

• Set default value for weight_sparsity in ResponseSelector to 0. This fixes a bug in the default behavior of ResponseSelector which was accidentally introduced in rasa==1.8.0. Users should update to this version and re-train their models if ResponseSelector was used in their pipeline.

[1.9.2] - 2020-03-26

Improved Documentation

• Fix documentation to bring back Sara.

[1.9.1] - 2020-03-25

Bugfixes

• Fix an issue where the deprecated queue parameter for the Pika Event Broker was ignored and Rasa Open Source published the events to the rasa_core_events queue instead. Note that this does not change the fact that the queue argument is deprecated in favor of the queues argument.

[1.9.0] - 2020-03-24

Features

• Channel hangouts for Rasa integration with Google Hangouts Chat is now supported out-of-the-box.

• Add an optional path to a specific directory to download and cache the pre-trained model weights for HFTransformersNLP.

• Add options tensorboard_log_directory and tensorboard_log_level to EmbeddingIntentClassifier, DIETClasifier, ResponseSelector, EmbeddingPolicy and TEDPolicy.

  By default tensorboard_log_directory is None. If a valid directory is provided, metrics are written during training. After the model is trained you can take a look at the training metrics in tensorboard. Execute tensorboard --logdir <path-to-given-directory>.

  Metrics can either be written after every epoch (default) or for every training step. You can specify when to write metrics using the variable tensorboard_log_level. Valid values are 'epoch' and 'minibatch'.

  We also write down a model summary, i.e. layers with inputs and types, to the given directory.

Improvements

• Make response timeout configurable. rasa run, rasa shell and rasa x can now be started with --response-timeout <int> to configure a response timeout of <int> seconds.

• Add full retrieval intent name to message data ResponseSelector will now add the full retrieval intent name e.g. faq/which_version to the prediction, making it accessible from the tracker.

• Added PikaEventBroker (Pika Event Broker) support for publishing to multiple queues. Messages are now published to a fanout exchange with name rasa-exchange (see exchange-fanout for more information on fanout exchanges).

  The former queue key is deprecated. Queues should now be specified as a list in the endpoints.yml event broker config under a new key queues. Example config:

  event_broker:
    type: pika
    url: localhost
    username: username
    password: password
    queues:
      - queue-1
      - queue-2
      - queue-3

• Change rasa init to include tests/conversation_tests.md file by default.

• The endpoint PUT /conversations/<conversation_id>/tracker/events no longer adds session start events (to learn more about conversation sessions, please see Session configuration) in addition to the events which were sent in the request payload. To achieve the old behavior send a GET /conversations/<conversation_id>/tracker request before appending events.

• Make scale_loss for intents behave the same way as in versions below 1.8, but only scale if some of the examples in a batch has probability of the golden label more than 0.5. Introduce scale_loss for entities in DIETClassifier.

Bugfixes

• Fixed the bug when FormPolicy was overwriting MappingPolicy prediction (e.g. /restart). Priorities for Mapping Policy and Form Policy are no longer linear: FormPolicy priority is 5, but its prediction is ignored if MappingPolicy is used for prediction.

• Fixed issue related to storing Python float values as decimal.Decimal objects in DynamoDB tracker stores. All decimal.Decimal objects are now converted to float on tracker retrieval.

  Added a new docs section on DynamoTrackerStore.

• Fixed bug where FallbackPolicy would always fall back if the fallback action is action_listen.

• Fixed bug where starting or ending a response with \\n\\n led to one of the responses returned being empty.

• Fixes issue where model always gets retrained if multiple NLU/story files are in a directory, by sorting the list of files.

• Fixed ambiguous logging in DIETClassifier by adding the name of the calling class to the log message.

Improved Documentation

• Restructure the “Evaluating models” documentation page and rename this page to Testing Your Assistant.

• Improved documentation on how to build and deploy an action server image for use on other servers such as Rasa X deployments.

Miscellaneous internal changes

Miscellaneous internal changes.

[1.8.3] - 2020-03-27

Bugfixes

• Fixes issue where model always gets retrained if multiple NLU/story files are in a directory, by sorting the list of files.

• Fixed ambiguous logging in DIETClassifier by adding the name of the calling class to the log message.

• Set default value for weight_sparsity in ResponseSelector to 0. This fixes a bug in the default behavior of ResponseSelector which was accidentally introduced in rasa==1.8.0. Users should update to this version or rasa>=1.9.3 and re-train their models if ResponseSelector was used in their pipeline.

Improved Documentation

• Improved documentation on how to build and deploy an action server image for use on other servers such as Rasa X deployments.

[1.8.2] - 2020-03-19

Bugfixes

• Fixed bug when installing rasa with poetry.

• Fixed bug with EmbeddingIntentClassifier, where results weren't the same as in 1.7.x. Fixed by setting weight sparsity to 0.

Improved Documentation

• Explain how to run commands as root user in Rasa SDK Docker images since version 1.8.0. Since version 1.8.0 the Rasa SDK Docker images does not longer run as root user by default. For commands which require root user usage, you have to switch back to the root user in your Docker image as described in Building an Action Server Image.

• Made improvements to Building Assistants tutorial

[1.8.1] - 2020-03-06

Bugfixes

• Fixed issue with using language models like xlnet along with entity_recognition set to True inside DIETClassifier.

Miscellaneous internal changes

Miscellaneous internal changes.

[1.8.0] - 2020-02-26

Deprecations and Removals

• Removed Agent.continue_training and the dump_flattened_stories parameter from Agent.persist.

• Properties Component.provides and Component.requires are deprecated. Use Component.required_components() instead.

Features

• Add default value __other__ to values of a CategoricalSlot.

  All values not mentioned in the list of values of a CategoricalSlot will be mapped to __other__ for featurization.

• Add story structure validation functionality (e.g. rasa data validate stories –max-history 5).

• Add LexicalSyntacticFeaturizer to sparse featurizers.

  LexicalSyntacticFeaturizer does the same featurization as the CRFEntityExtractor. We extracted the featurization into a separate component so that the features can be reused and featurization is independent from the entity extraction.

• Integrate language models from HuggingFace's Transformers Library.

  Add a new NLP component HFTransformersNLP which tokenizes and featurizes incoming messages using a specified pre-trained model with the Transformers library as the backend. Add LanguageModelTokenizer and LanguageModelFeaturizer which use the information from HFTransformersNLP and sets them correctly for message object. Language models currently supported: BERT, OpenAIGPT, GPT-2, XLNet, DistilBert, RoBERTa.

• Added a new CLI command rasa export to publish tracker events from a persistent tracker store using an event broker. See Export Conversations to an Event Broker, Tracker Stores and Event Brokers for more details.

• Refactor how GPU and CPU environments are configured for TensorFlow 2.0.

  Please refer to the documentation on Configuring TensorFlow to understand which environment variables to set in what scenarios. A couple of examples are shown below as well:

  # This specifies to use 1024 MB of memory from GPU with logical ID 0 and 2048 MB of memory from GPU with logical ID 1
  TF_GPU_MEMORY_ALLOC="0:1024, 1:2048"
  
  # Specifies that at most 3 CPU threads can be used to parallelize multiple non-blocking operations
  TF_INTER_OP_PARALLELISM_THREADS="3"
  
  # Specifies that at most 2 CPU threads can be used to parallelize a particular operation.
  TF_INTRA_OP_PARALLELISM_THREADS="2"

• Added a new NLU component DIETClassifier and a new policy TEDPolicy.

  DIET (Dual Intent and Entity Transformer) is a multi-task architecture for intent classification and entity recognition. You can read more about this component in the DIETClassifier documentation. The new component will replace the EmbeddingIntentClassifier and the CRFEntityExtractor in the future. Those two components are deprecated from now on. See migration guide for details on how to switch to the new component.

  TEDPolicy is the new name for EmbeddingPolicy. EmbeddingPolicy is deprecated from now on. The functionality of TEDPolicy and EmbeddingPolicy is the same. Please update your configuration file to use the new name for the policy.

• The sentence vector of the SpacyFeaturizer and MitieFeaturizer can be calculated using max or mean pooling.

  To specify the pooling operation, set the option pooling for the SpacyFeaturizer or the MitieFeaturizer in your configuration file. The default pooling operation is mean. The mean pooling operation also does not take into account words, that do not have a word vector.

Improvements

• Added command line argument --conversation-id to rasa interactive. If the argument is not given, conversation_id defaults to a random uuid.

• Added a new command-line argument --init-dir to command rasa init to specify the directory in which the project is initialised.

• Added support to send images with the twilio output channel.

• Part of Slack sanitization: Multiple garbled URL's in a string coming from slack will be converted into actual strings. Example: health check of <http://eemdb.net|eemdb.net> and <http://eemdb1.net|eemdb1.net> to health check of eemdb.net and eemdb1.net

• New command-line argument –conversation-id will be added and wiil give the ability to set specific conversation ID for each shell session, if not passed will be random.

• Messages sent to the Pika Event Broker are now persisted. This guarantees the RabbitMQ will re-send previously received messages after a crash. Note that this does not help for the case where messages are sent to an unavailable RabbitMQ instance.

• Added support for mattermost connector to use bot accounts.

• We updated our code to TensorFlow 2.

• Events exported using rasa export receive a message header if published through a PikaEventBroker. The header is added to the message's BasicProperties.headers under the rasa-export-process-id key (rasa.core.constants.RASA_EXPORT_PROCESS_ID_HEADER_NAME). The value is a UUID4 generated at each call of rasa export. The resulting header is a key-value pair that looks as follows:

  'rasa-export-process-id': 'd3b3d3ffe2bd4f379ccf21214ccfb261'

• Added followlinks=True to os.walk calls, to allow the use of symlinks in training, NLU and domain data.

• Support invoking a SlackBot by direct messaging or @<app name> mentions.

Bugfixes

• Fixed timestamp parsing warning when using DucklingHTTPExtractor

• Fixed issue with action_restart getting overridden by action_listen when the Mapping Policy and the Two-Stage Fallback Policy are used together.

• Fixed incorrectly raised Error encountered in pipelines with a ResponseSelector and NLG.

  When NLU training data is split before NLU pipeline comparison, NLG responses were not also persisted and therefore training for a pipeline including the ResponseSelector would fail.

  NLG responses are now persisted along with NLU data to a /train directory in the run_x/xx%_exclusion folder.

• Fixed sending custom json with Twilio channel

Improved Documentation

• Updated the documentation to properly suggest not to explicitly add utterance actions to the domain.

• Added user guide for reminders and external events, including reminderbot demo.

Miscellaneous internal changes

Miscellaneous internal changes.

[1.7.4] - 2020-02-24

Bugfixes

• Tracker stores supporting conversation sessions (SQLTrackerStore and MongoTrackerStore) do not save the tracker state to database immediately after starting a new conversation session. This leads to the number of events being saved in addition to the already-existing ones to be calculated correctly.

  This fixes action_listen events being saved twice at the beginning of conversation sessions.

[1.7.3] - 2020-02-21

Bugfixes

• Fix segmentation fault when running rasa train or rasa shell.

Improved Documentation

• Fix doc links on “Deploying your Assistant” page

[1.7.2] - 2020-02-13

Bugfixes

• Fixed incompatibility of Oracle with the SQLTrackerStore, by using a Sequence for the primary key columns. This does not change anything for SQL databases other than Oracle. If you are using Oracle, please create a sequence with the instructions in the SQLTrackerStore docs.

Improved Documentation

• Added section on setting up the SQLTrackerStore with Oracle

• Renamed “Running the Server” page to “Configuring the HTTP API”

[1.7.1] - 2020-02-11

Bugfixes

• Fixed file loading of non proper UTF-8 story files, failing properly when checking for story files.

• Fix problem with multi-intents. Training with multi-intents using the CountVectorsFeaturizer together with EmbeddingIntentClassifier is working again.

• Fix bug ValueError: Cannot concatenate sparse features as sequence dimension does not match.

  When training a Rasa model that contains responses for just some of the intents, training was failing. Fixed the featurizers to return a consistent feature vector in case no response was given for a specific message.

• If no text features are present in EmbeddingIntentClassifier return the intent None.

• Resolve version conflicts: Pin version of cloudpickle to ~=1.2.0.

[1.7.0] - 2020-01-29

Deprecations and Removals

• The endpoint /conversations/<conversation_id>/execute is now deprecated. Instead, users should use the /conversations/<conversation_id>/trigger_intent endpoint and thus trigger intents instead of actions.

• Remove option use_cls_token from tokenizers and option return_sequence from featurizers.

  By default all tokenizer add a special token (__CLS__) to the end of the list of tokens. This token will be used to capture the features of the whole utterance.

  The featurizers will return a matrix of size (number-of-tokens x feature-dimension) by default. This allows to train sequence models. However, the feature vector of the __CLS__ token can be used to train non-sequence models. The corresponding classifier can decide what kind of features to use.

Features

• Rename templates key in domain to responses.

  templates key will still work for backwards compatibility but will raise a future warning.

• Added a new configuration parameter, ranking_length to the EmbeddingPolicy, EmbeddingIntentClassifier, and ResponseSelector classes.

• External events and reminders now trigger intents (and entities) instead of actions.

  Add new endpoint /conversations/<conversation_id>/trigger_intent, which lets the user specify an intent and a list of entities that is injected into the conversation in place of a user message. The bot then predicts and executes a response action.

• Add ConveRTTokenizer.

  The tokenizer should be used whenever the ConveRTFeaturizer is used.

  Every tokenizer now supports the following configuration options: intent_tokenization_flag: Flag to check whether to split intents (default False). intent_split_symbol: Symbol on which intent should be split (default _)

Improvements

• Remove the need of specifying utter actions in the actions section explicitly if these actions are already listed in the templates section.

• Entity examples that have been extracted using an external extractor are excluded from Markdown dumping in MarkdownWriter.dumps(). The excluded external extractors are DucklingHTTPExtractor and SpacyEntityExtractor.

• The EmbeddingPolicy, EmbeddingIntentClassifier, and ResponseSelector now by default normalize confidence levels over the top 10 results. See Rasa 1.6 to Rasa 1.7 for more details.

• ReminderCancelled can now cancel multiple reminders if no name is given. It still cancels a single reminder if the reminder's name is specified.

Bugfixes

• Requests to /model/train do not longer block other requests to the Rasa server.

• Fixed default behavior of rasa test core --evaluate-model-directory when called without --model. Previously, the latest model file was used as --model. Now the default model directory is used instead.

  New behavior of rasa test core --evaluate-model-directory when given an existing file as argument for --model: Previously, this led to an error. Now a warning is displayed and the directory containing the given file is used as --model.

• Updated the dependency networkx from 2.3.0 to 2.4.0. The old version created incompatibilities when using pip.

  There is an imcompatibility between Rasa dependecy requests 2.22.0 and the own depedency from Rasa for networkx raising errors upon pip install. There is also a bug corrected in requirements.txt which used ~= instead of ==. All of these are fixed using networkx 2.4.0.

• Fixed compatibility issue with Microsoft Bot Framework Emulator if service_url lacked a trailing /.

• DynamoDB tracker store decimal values will now be rounded on save. Previously values exceeding 38 digits caused an unhandled error.

Miscellaneous internal changes

Miscellaneous internal changes.

[1.6.2] - 2020-01-28

Improvements

• Switching back to a TensorFlow release which only includes CPU support to reduce the size of the dependencies. If you want to use the TensorFlow package with GPU support, please run pip install tensorflow-gpu==1.15.0.

Bugfixes

• Fixes Exception 'Loop' object has no attribute '_ready' error when running rasa init.

• Updated the end-to-end ValueError you recieve when you have a invalid story format to point to the updated doc link.

[1.6.1] - 2020-01-07

Bugfixes

• Use an empty domain in case a model is loaded which has no domain (avoids errors when accessing agent.doman.<some attribute>).

• Replace error message with warning in tokenizers and featurizers if default parameter not set.

• Pin sanic patch version instead of minor version. Fixes sanic _run_request_middleware() error.

• Fix wrong calculation of additional conversation events when saving the conversation. This led to conversation events not being saved.

• Fix wrong order of conversation events when pushing events to conversations via POST /conversations/<conversation_id>/tracker/events.

[1.6.0] - 2019-12-18

Deprecations and Removals

• Removed ner_features as a feature name from CRFEntityExtractor, use text_dense_features instead.

  The following settings match the previous NGramFeaturizer:

  pipeline:
  - name: 'CountVectorsFeaturizer'
    analyzer: 'char_wb'
    min_ngram: 3
    max_ngram: 17
    max_features: 10
    min_df: 5

• To use custom features in the CRFEntityExtractor use text_dense_features instead of ner_features. If text_dense_features are present in the feature set, the CRFEntityExtractor will automatically make use of them. Just make sure to add a dense featurizer in front of the CRFEntityExtractor in your pipeline and set the flag return_sequence to True for that featurizer.

• Deprecated Agent.continue_training. Instead, a model should be retrained.

• Specifying lookup tables directly in the NLU file is now deprecated. Please specify them in an external file.

Features

• Replaced the warnings about missing templates, intents etc. in validator.py by debug messages.

• Added conversation sessions to trackers.

  A conversation session represents the dialog between the assistant and a user. Conversation sessions can begin in three ways: 1. the user begins the conversation with the assistant, 2. the user sends their first message after a configurable period of inactivity, or 3. a manual session start is triggered with the /session_start intent message. The period of inactivity after which a new conversation session is triggered is defined in the domain using the session_expiration_time key in the session_config section. The introduction of conversation sessions comprises the following changes:

  • Added a new event SessionStarted that marks the beginning of a new conversation session.

  • Added a new default action ActionSessionStart. This action takes all SlotSet events from the previous session and applies it to the next session.

  • Added a new default intent session_start which triggers the start of a new conversation session.

  • SQLTrackerStore and MongoTrackerStore only retrieve events from the last session from the database.

  note

  The session behavior is disabled for existing projects, i.e. existing domains without session config section.

• Preparation for an upcoming change in the EmbeddingIntentClassifier:

  Add option use_cls_token to all tokenizers. If it is set to True, the token __CLS__ will be added to the end of the list of tokens. Default is set to False. No need to change the default value for now.

  Add option return_sequence to all featurizers. By default all featurizers return a matrix of size (1 x feature-dimension). If the option return_sequence is set to True, the corresponding featurizer will return a matrix of size (token-length x feature-dimension). See Text Featurizers. Default value is set to False. However, you might want to set it to True if you want to use custom features in the CRFEntityExtractor. See passing custom features to the CRFEntityExtractor

  Changed some featurizers to use sparse features, which should reduce memory usage with large amounts of training data significantly. Read more: Text Featurizers .

  caution

  These changes break model compatibility. You will need to retrain your old models!

Improvements

• Added --no-plot option for rasa test command, which disables rendering of confusion matrix and histogram. By default plots will be rendered.

• If matplotlib couldn't set up a default backend, it will be set automatically to TkAgg/Agg one

• Add the option \random_seed`to the`rasa data split nlu`` command to generate reproducible train/test splits.

• Changed url __init__() arguments for custom tracker stores to host to reflect the __init__ arguments of currently supported tracker stores. Note that in endpoints.yml, these are still declared as url.

• The kafka-python dependency has become as an “extra” dependency. To use the KafkaEventConsumer, rasa has to be installed with the [kafka] option, i.e.

  $ pip install rasa[kafka]

• Allow creation of natural language interpreter and generator by classname reference in endpoints.yml.

• Made it explicit that interactive learning does not work with NLU-only models.

  Interactive learning no longer trains NLU-only models if no model is provided and no core data is provided.

• The intent_report.json created by rasa test now creates an extra field confused_with for each intent. This is a dictionary containing the names of the most common false positives when this intent should be predicted, and the number of such false positives.

• rasa test nlu --cross-validation now also includes an evaluation of the response selector. As a result, the train and test F1-score, accuracy and precision is logged for the response selector. A report is also generated in the results folder by the name response_selection_report.json

Bugfixes

• If a wait_time_between_pulls is configured for the model server in endpoints.yml, this will be used instead of the default one when running Rasa X.

• Training Luis data with luis_schema_version higher than 4.x.x will show a warning instead of throwing an exception.

• Running rasa interactive with no NLU data now works, with the functionality of rasa interactive core.

• When loading models from S3, namespaces (folders within a bucket) are now respected. Previously, this would result in an error upon loading the model.

• “rasa init” will ask if user wants to train a model

• Pin multidict dependency to 4.6.1 to prevent sanic from breaking, see the Sanic GitHub issue for more info.

• Fix errors during training and testing of ResponseSelector.

[1.5.3] - 2019-12-11

Improvements

• Improved error message that appears when an incorrect parameter is passed to a policy.

Bugfixes

• Added rasa/nlu/schemas/config.yml to wheel package

• Pin multidict dependency to 4.6.1 to prevent sanic from breaking, see the Sanic GitHub issue

[1.5.2] - 2019-12-09

Improvements

• rasa interactive will skip the story visualization of training stories in case there are more than 200 stories. Stories created during interactive learning will be visualized as before.

• The log level for SocketIO loggers, including websockets.protocol, engineio.server, and socketio.server, is now handled by the LOG_LEVEL_LIBRARIES environment variable, where the default log level is ERROR.

• Updated all example bots and documentation to use the updated dispatcher.utter_message() method from rasa-sdk==1.5.0.

Bugfixes

• rasa interactive will not load training stories in case the visualization is skipped.

• Fixed error where spacy models where not found in the docker images.

• Fixed unnecessary kwargs unpacking in rasa.test.test_core call in rasa.test.test function.

• Training data files now get loaded in the same order (especially relevant to subdirectories) each time to ensure training consistency when using a random seed.

• Locks for tickets in LockStore are immediately issued without a redundant check for their availability.

Improved Documentation

• Added towncrier to automatically collect changelog entries.

• Document the pipeline for pretrained_embeddings_convert in the pre-configured pipelines section.

• Proactively Reaching Out to the User Using Actions now correctly links to the endpoint specification.

[1.5.1] - 2019-11-27

Improvements

• When NLU training data is dumped as Markdown file the intents are not longer ordered alphabetically, but in the original order of given training data

Bugfixes

• End to end stories now support literal payloads which specify entities, e.g. greet: /greet{"name": "John"}

• Slots will be correctly interpolated if there are lists in custom response templates.

• Fixed compatibility issues with rasa-sdk 1.5

• Updated /status endpoint to show correct path to model archive

[1.5.0] - 2019-11-26

Features

• Added data validator that checks if domain object returned is empty. If so, exit early from the command rasa data validate.

• Added the KeywordIntentClassifier.

• Added documentation for AugmentedMemoizationPolicy.

• Fall back to InMemoryTrackerStore in case there is any problem with the current tracker store.

• Arbitrary metadata can now be attached to any Event subclass. The data must be stored under the metadata key when reading the event from a JSON object or dictionary.

• Add command line argument rasa x --config CONFIG, to specify path to the policy and NLU pipeline configuration of your bot (default: config.yml).

• Added a new NLU featurizer - ConveRTFeaturizer based on ConveRT model released by PolyAI.

• Added a new preconfigured pipeline - pretrained_embeddings_convert.

Improvements

• Do not retrain the entire Core model if only the templates section of the domain is changed.

• Upgraded jsonschema version.

Deprecations and Removals

• Remove duplicate messages when creating training data (issues/1446).

Bugfixes

• MultiProjectImporter now imports files in the order of the import statements

• Fixed server hanging forever on leaving rasa shell before first message

• Fixed rasa init showing traceback error when user does Keyboard Interrupt before choosing a project path

• CountVectorsFeaturizer featurizes intents only if its analyzer is set to word

• Fixed bug where facebooks generic template was not rendered when buttons were None

• Fixed default intents unnecessarily raising undefined parsing error

[1.4.6] - 2019-11-22

Bugfixes

• Fixed Rasa X not working when any tracker store was configured for Rasa.

• Use the matplotlib backend agg in case the tkinter package is not installed.

[1.4.5] - 2019-11-14

Bugfixes

• NLU-only models no longer throw warnings about parsing features not defined in the domain

• Fixed bug that stopped Dockerfiles from building version 1.4.4.

• Fixed format guessing for e2e stories with intent restated as /intent

[1.4.4] - 2019-11-13

Features

• PikaEventProducer adds the RabbitMQ App ID message property to published messages with the value of the RASA_ENVIRONMENT environment variable. The message property will not be assigned if this environment variable isn't set.

Improvements

• Updated Mattermost connector documentation to be more clear.

• Updated format strings to f-strings where appropriate.

• Updated tensorflow requirement to 1.15.0

• Dump domain using UTF-8 (to avoid \\UXXXX sequences in the dumped files)

Bugfixes

• Fixed exporting NLU training data in json format from rasa interactive

• Fixed numpy deprecation warnings

[1.4.3] - 2019-10-29

Bugfixes

• Fixed Connection reset by peer errors and bot response delays when using the RabbitMQ event broker.

[1.4.2] - 2019-10-28

Deprecations and Removals

• TensorFlow deprecation warnings are no longer shown when running rasa x

Bugfixes

• Fixed 'Namespace' object has no attribute 'persist_nlu_data' error during interactive learning

• Pinned networkx~=2.3.0 to fix visualization in rasa interactive and Rasa X

• Fixed No model found error when using rasa run actions with “actions” as a directory.

[1.4.1] - 2019-10-22

Regression: changes from 1.2.12 were missing from 1.4.0, readded them

[1.4.0] - 2019-10-19

Features

• add flag to CLI to persist NLU training data if needed

• log a warning if the Interpreter picks up an intent or an entity that does not exist in the domain file.

• added DynamoTrackerStore to support persistence of agents running on AWS

• added docstrings for TrackerStore classes

• added buttons and images to mattermost.

• CRFEntityExtractor updated to accept arbitrary token-level features like word vectors (issues/4214)

• SpacyFeaturizer updated to add ner_features for CRFEntityExtractor

• Sanitizing incoming messages from slack to remove slack formatting like <mailto:xyz@rasa.com|xyz@rasa.com> or <http://url.com|url.com> and substitute it with original content

• Added the ability to configure the number of Sanic worker processes in the HTTP server (rasa.server) and input channel server (rasa.core.agent.handle_channels()). The number of workers can be set using the environment variable SANIC_WORKERS (default: 1). A value of >1 is allowed only in combination with RedisLockStore as the lock store.

• Botframework channel can handle uploaded files in UserMessage metadata.

• Added data validator that checks there is no duplicated example data across multiples intents

Improvements

• Unknown sections in markdown format (NLU data) are not ignored anymore, but instead an error is raised.

• It is now easier to add metadata to a UserMessage in existing channels. You can do so by overwriting the method get_metadata. The return value of this method will be passed to the UserMessage object.

• Tests can now be run in parallel

• Serialise DialogueStateTracker as json instead of pickle. DEPRECATION warning: Deserialisation of pickled trackers will be deprecated in version 2.0. For now, trackers are still loaded from pickle but will be dumped as json in any subsequent save operations.

• Event brokers are now also passed to custom tracker stores (using the event_broker parameter)

• Don't run the Rasa Docker image as root.

• Use multi-stage builds to reduce the size of the Rasa Docker image.

• Updated the /status api route to use the actual model file location instead of the tmp location.

Deprecations and Removals

• Removed Python 3.5 support

Bugfixes

• fixed missing tkinter dependency for running tests on Ubuntu

• fixed issue with conversation JSON serialization

• fixed the hanging HTTP call with ner_duckling_http pipeline

• fixed Interactive Learning intent payload messages saving in nlu files

• fixed DucklingHTTPExtractor dimensions by actually applying to the request

[1.3.10] - 2019-10-18

Features

• Can now pass a package as an argument to the --actions parameter of the rasa run actions command.

Bugfixes

• Fixed visualization of stories with entities which led to a failing visualization in Rasa X

[1.3.9] - 2019-10-10

Features

• Port of 1.2.10 (support for RabbitMQ TLS authentication and port key in event broker endpoint config).

• Port of 1.2.11 (support for passing a CA file for SSL certificate verification via the –ssl-ca-file flag).

Bugfixes

• Fixed the hanging HTTP call with ner_duckling_http pipeline.

• Fixed text processing of intent attribute inside CountVectorFeaturizer.

• Fixed argument of type 'NoneType' is not iterable when using rasa shell, rasa interactive / rasa run

[1.3.8] - 2019-10-08

Improvements

• Policies now only get imported if they are actually used. This removes TensorFlow warnings when starting Rasa X

Bugfixes

• Fixed error Object of type 'MaxHistoryTrackerFeaturizer' is not JSON serializable when running rasa train core

• Default channel send_ methods no longer support kwargs as they caused issues in incompatible channels

[1.3.7] - 2019-09-27

Bugfixes

• re-added TLS, SRV dependencies for PyMongo

• socketio can now be run without turning on the --enable-api flag

• MappingPolicy no longer fails when the latest action doesn't have a policy

[1.3.6] - 2019-09-21

Features

• Added the ability for users to specify a conversation id to send a message to when using the RasaChat input channel.

[1.3.5] - 2019-09-20

Bugfixes

• Fixed issue where rasa init would fail without spaCy being installed

[1.3.4] - 2019-09-20

Features

• Added the ability to set the backlog parameter in Sanics run() method using the SANIC_BACKLOG environment variable. This parameter sets the number of unaccepted connections the server allows before refusing new connections. A default value of 100 is used if the variable is not set.

• Status endpoint (/status) now also returns the number of training processes currently running

Bugfixes

• Added the ability to properly deal with spaCy Doc-objects created on empty strings as discussed in issue #4445. Only training samples that actually bear content are sent to self.nlp.pipe for every given attribute. Non-content-bearing samples are converted to empty Doc-objects. The resulting lists are merged with their preserved order and properly returned.

• asyncio warnings are now only printed if the callback takes more than 100ms (up from 1ms).

• agent.load_model_from_server no longer affects logging.

Improvements

• The endpoint POST /model/train no longer supports specifying an output directory for the trained model using the field out. Instead you can choose whether you want to save the trained model in the default model directory (models) (default behavior) or in a temporary directory by specifying the save_to_default_model_directory field in the training request.

[1.3.3] - 2019-09-13

Bugfixes

• Added a check to avoid training CountVectorizer for a particular attribute of a message if no text is provided for that attribute across the training data.

• Default one-hot representation for label featurization inside EmbeddingIntentClassifier if label features don't exist.

• Policy ensemble no longer incorrectly wrings “missing mapping policy” when mapping policy is present.

• “text” from utter_custom_json now correctly saved to tracker when using telegram channel

Deprecations and Removals

• Removed computation of intent_spacy_doc. As a result, none of the spacy components process intents now.

[1.3.2] - 2019-09-10

Bugfixes

• SQL tracker events are retrieved ordered by timestamps. This fixes interactive learning events being shown in the wrong order.

[1.3.1] - 2019-09-09

Improvements

• Pin gast to == 0.2.2

[1.3.0] - 2019-09-05

Features

• Added option to persist nlu training data (default: False)

• option to save stories in e2e format for interactive learning

• bot messages contain the timestamp of the BotUttered event, which can be used in channels

• FallbackPolicy can now be configured to trigger when the difference between confidences of two predicted intents is too narrow

• experimental training data importer which supports training with data of multiple sub bots. Please see the docs for more information.

• throw error during training when triggers are defined in the domain without MappingPolicy being present in the policy ensemble

• The tracker is now available within the interpreter's parse method, giving the ability to create interpreter classes that use the tracker state (eg. slot values) during the parsing of the message. More details on motivation of this change see issues/3015.

• add example bot knowledgebasebot to showcase the usage of ActionQueryKnowledgeBase

• softmax starspace loss for both EmbeddingPolicy and EmbeddingIntentClassifier

• balanced batching strategy for both EmbeddingPolicy and EmbeddingIntentClassifier

• max_history parameter for EmbeddingPolicy

• Successful predictions of the NER are written to a file if --successes is set when running rasa test nlu

• Incorrect predictions of the NER are written to a file by default. You can disable it via --no-errors.

• New NLU component ResponseSelector added for the task of response selection

• Message data attribute can contain two more keys - response_key, response depending on the training data

• New action type implemented by ActionRetrieveResponse class and identified with response_ prefix

• Vocabulary sharing inside CountVectorsFeaturizer with use_shared_vocab flag. If set to True, vocabulary of corpus is shared between text, intent and response attributes of message

• Added an option to share the hidden layer weights of text input and label input inside EmbeddingIntentClassifier using the flag share_hidden_layers

• New type of training data file in NLU which stores response phrases for response selection task.

• Add flag intent_split_symbol and intent_tokenization_flag to all WhitespaceTokenizer, JiebaTokenizer and SpacyTokenizer

• Added evaluation for response selector. Creates a report response_selection_report.json inside --out directory.

• argument --config-endpoint to specify the URL from which rasa x pulls the runtime configuration (endpoints and credentials)

• LockStore class storing instances of TicketLock for every conversation_id

• environment variables SQL_POOL_SIZE (default: 50) and SQL_MAX_OVERFLOW (default: 100) can be set to control the pool size and maximum pool overflow for SQLTrackerStore when used with the postgresql dialect

• Add a bot_challenge intent and a utter_iamabot action to all example projects and the rasa init bot.

• Allow sending attachments when using the socketio channel

• rasa data validate will fail with a non-zero exit code if validation fails

Improvements

• added character-level CountVectorsFeaturizer with empirically found parameters into the supervised_embeddings NLU pipeline template

• NLU evaluations now also stores its output in the output directory like the core evaluation

• show warning in case a default path is used instead of a provided, invalid path

• compare mode of rasa train core allows the whole core config comparison, naming style of models trained for comparison is changed (this is a breaking change)

• pika keeps a single connection open, instead of open and closing on each incoming event

• RasaChatInput fetches the public key from the Rasa X API. The key is used to decode the bearer token containing the conversation ID. This requires rasa-x>=0.20.2.

• more specific exception message when loading custom components depending on whether component's path or class name is invalid or can't be found in the global namespace

• change priorities so that the MemoizationPolicy has higher priority than the MappingPolicy

• substitute LSTM with Transformer in EmbeddingPolicy

• EmbeddingPolicy can now use MaxHistoryTrackerFeaturizer

• non zero evaluate_on_num_examples in EmbeddingPolicy and EmbeddingIntentClassifier is the size of hold out validation set that is excluded from training data

• defaults parameters and architectures for both EmbeddingPolicy and EmbeddingIntentClassifier are changed (this is a breaking change)

• evaluation of NER does not include 'no-entity' anymore

• --successes for rasa test nlu is now boolean values. If set incorrect/successful predictions are saved in a file.

• --errors is renamed to --no-errors and is now a boolean value. By default incorrect predictions are saved in a file. If --no-errors is set predictions are not written to a file.

• Remove label_tokenization_flag and label_split_symbol from EmbeddingIntentClassifier. Instead move these parameters to Tokenizers.

• Process features of all attributes of a message, i.e. - text, intent and response inside the respective component itself. For e.g. - intent of a message is now tokenized inside the tokenizer itself.

• Deprecate as_markdown and as_json in favour of nlu_as_markdown and nlu_as_json respectively.

• pin python-engineio >= 3.9.3

• update python-socketio req to >= 4.3.1

Bugfixes

• rasa test nlu with a folder of configuration files

• MappingPolicy standard featurizer is set to None

• Removed text parameter from send_attachment function in slack.py to avoid duplication of text output to slackbot

• server /status endpoint reports status when an NLU-only model is loaded

Deprecations and Removals

• Removed --report argument from rasa test nlu. All output files are stored in the --out directory.

[1.2.12] - 2019-10-16

Features

• Support for transit encryption with Redis via use_ssl: True in the tracker store config in endpoints.yml

[1.2.11] - 2019-10-09

Features

• Support for passing a CA file for SSL certificate verification via the –ssl-ca-file flag

[1.2.10] - 2019-10-08

Features

• Added support for RabbitMQ TLS authentication. The following environment variables need to be set: RABBITMQ_SSL_CLIENT_CERTIFICATE - path to the SSL client certificate (required) RABBITMQ_SSL_CLIENT_KEY - path to the SSL client key (required) RABBITMQ_SSL_CA_FILE - path to the SSL CA file (optional, for certificate verification) RABBITMQ_SSL_KEY_PASSWORD - SSL private key password (optional)

• Added ability to define the RabbitMQ port using the port key in the event_broker endpoint config.

[1.2.9] - 2019-09-17

Bugfixes

• Correctly pass SSL flag values to x CLI command (backport of

[1.2.8] - 2019-09-10

Bugfixes

• SQL tracker events are retrieved ordered by timestamps. This fixes interactive learning events being shown in the wrong order. Backport of 1.3.2 patch (PR #4427).

[1.2.7] - 2019-09-02

Bugfixes

• Added query dictionary argument to SQLTrackerStore which will be appended to the SQL connection URL as query parameters.

[1.2.6] - 2019-09-02

Bugfixes

• fixed bug that occurred when sending template elements through a channel that doesn't support them

[1.2.5] - 2019-08-26

Features

• SSL support for rasa run command. Certificate can be specified using --ssl-certificate and --ssl-keyfile.

Bugfixes

• made default augmentation value consistent across repo

• '/restart' will now also restart the bot if the tracker is paused

[1.2.4] - 2019-08-23

Bugfixes

• the SocketIO input channel now allows accesses from other origins (fixes SocketIO channel on Rasa X)

[1.2.3] - 2019-08-15

Improvements

• messages with multiple entities are now handled properly with e2e evaluation

• data/test_evaluations/end_to_end_story.md was re-written in the restaurantbot domain

[1.2.3] - 2019-08-15

Improvements

• messages with multiple entities are now handled properly with e2e evaluation

• data/test_evaluations/end_to_end_story.md was re-written in the restaurantbot domain

Bugfixes

• Free text input was not allowed in the Rasa shell when the response template contained buttons, which has now been fixed.

[1.2.2] - 2019-08-07

Bugfixes

• UserUttered events always got the same timestamp

[1.2.1] - 2019-08-06

Features

• Docs now have an EDIT THIS PAGE button

Bugfixes

• Flood control exceeded error in Telegram connector which happened because the webhook was set twice

[1.2.0] - 2019-08-01

Features

• add root route to server started without --enable-api parameter

• add --evaluate-model-directory to rasa test core to evaluate models from rasa train core -c <config-1> <config-2>

• option to send messages to the user by calling POST /conversations/\{conversation_id\}/execute

Improvements

• Agent.update_model() and Agent.handle_message() now work without needing to set a domain or a policy ensemble

• Update pytype to 2019.7.11

• new event broker class: SQLProducer. This event broker is now used when running locally with Rasa X

• API requests are not longer logged to rasa_core.log by default in order to avoid problems when running on OpenShift (use --log-file rasa_core.log to retain the old behavior)

• metadata attribute added to UserMessage

Bugfixes

• rasa test core can handle compressed model files

• rasa can handle story files containing multi line comments

• template will retain { if escaped with {. e.g. {{“foo”: {bar}}} will result in {“foo”: “replaced value”}

[1.1.8] - 2019-07-25

Features

• TrainingFileImporter interface to support customizing the process of loading training data

• fill slots for custom templates

Improvements

• Agent.update_model() and Agent.handle_message() now work without needing to set a domain or a policy ensemble

• update pytype to 2019.7.11

Bugfixes

• interactive learning bug where reverted user utterances were dumped to training data

• added timeout to terminal input channel to avoid freezing input in case of server errors

• fill slots for image, buttons, quick_replies and attachments in templates

• rasa train core in comparison mode stores the model files compressed (tar.gz files)

• slot setting in interactive learning with the TwoStageFallbackPolicy

[1.1.7] - 2019-07-18

Features

• added optional pymongo dependencies [tls, srv] to requirements.txt for better mongodb support

• case_sensitive option added to WhiteSpaceTokenizer with true as default.

Bugfixes

• validation no longer throws an error during interactive learning

• fixed wrong cleaning of use_entities in case it was a list and not True

• updated the server endpoint /model/parse to handle also messages with the intent prefix

• fixed bug where “No model found” message appeared after successfully running the bot

• debug logs now print to rasa_core.log when running rasa x -vv or rasa run -vv

[1.1.6] - 2019-07-12

Features

• rest channel supports setting a message's input_channel through a field input_channel in the request body

Improvements

• recommended syntax for empty use_entities and ignore_entities in the domain file has been updated from False or None to an empty list ([])

Bugfixes

• rasa run without --enable-api does not require a local model anymore

• using rasa run with --enable-api to run a server now prints “running Rasa server” instead of “running Rasa Core server”

• actions, intents, and utterances created in rasa interactive can no longer be empty

[1.1.5] - 2019-07-10

Features

• debug logging now tells you which tracker store is connected

• the response of /model/train now includes a response header for the trained model filename

• Validator class to help developing by checking if the files have any errors

• project's code is now linted using flake8

• info log when credentials were provided for multiple channels and channel in --connector argument was specified at the same time

• validate export paths in interactive learning

Improvements

• deprecate rasa.core.agent.handle_channels(...)\. Please use ``rasa.run(...)orrasa.core.run.configure_app` instead.

• Agent.load() also accepts tar.gz model file

Deprecations and Removals

• revert the stripping of trailing slashes in endpoint URLs since this can lead to problems in case the trailing slash is actually wanted

• starter packs were removed from Github and are therefore no longer tested by Travis script

Bugfixes

• all temporal model files are now deleted after stopping the Rasa server

• rasa shell nlu now outputs unicode characters instead of \\uxxxx codes

• fixed PUT /model with model_server by deserializing the model_server to EndpointConfig.

• x in AnySlotDict is now True for any x, which fixes empty slot warnings in interactive learning

• rasa train now also includes NLU files in other formats than the Rasa format

• rasa train core no longer crashes without a --domain arg

• rasa interactive now looks for endpoints in endpoints.yml if no --endpoints arg is passed

• custom files, e.g. custom components and channels, load correctly when using the command line interface

• MappingPolicy now works correctly when used as part of a PolicyEnsemble

[1.1.4] - 2019-06-18

Features

• unfeaturize single entities

• added agent readiness check to the /status resource

Improvements

• removed leading underscore from name of '_create_initial_project' function.

Bugfixes

• fixed bug where facebook quick replies were not rendering

• take FB quick reply payload rather than text as input

• fixed bug where training_data path in metadata.json was an absolute path

[1.1.3] - 2019-06-14

Bugfixes

• fixed any inconsistent type annotations in code and some bugs revealed by type checker

[1.1.2] - 2019-06-13

Bugfixes

• fixed duplicate events appearing in tracker when using a PostgreSQL tracker store

[1.1.1] - 2019-06-13

Bugfixes

• fixed compatibility with Rasa SDK

• bot responses can contain custom messages besides other message types

[1.1.0] - 2019-06-13

Features

• nlu configs can now be directly compared for performance on a dataset in rasa test nlu

Improvements

• update the tracker in interactive learning through reverting and appending events instead of replacing the tracker

• POST /conversations/\{conversation_id\}/tracker/events supports a list of events

Bugfixes

• fixed creation of RasaNLUHttpInterpreter

• form actions are included in domain warnings

• default actions, which are overriden by custom actions and are listed in the domain are excluded from domain warnings

• SQL data column type to Text for compatibility with MySQL

• non-featurizer training parameters don't break SklearnPolicy anymore

[1.0.9] - 2019-06-10

Improvements

• revert PR #3739 (as this is a breaking change): set PikaProducer and KafkaProducer default queues back to rasa_core_events

[1.0.8] - 2019-06-10

Features

• support for specifying full database urls in the SQLTrackerStore configuration

• maximum number of predictions can be set via the environment variable MAX_NUMBER_OF_PREDICTIONS (default is 10)

Improvements

• default PikaProducer and KafkaProducer queues to rasa_production_events

• exclude unfeaturized slots from domain warnings

Bugfixes

• loading of additional training data with the SkillSelector

• strip trailing slashes in endpoint URLs

[1.0.7] - 2019-06-06

Features

• added argument --rasa-x-port to specify the port of Rasa X when running Rasa X locally via rasa x

Bugfixes

• slack notifications from bots correctly render text

• fixed usage of --log-file argument for rasa run and rasa shell

• check if correct tracker store is configured in local mode

[1.0.6] - 2019-06-03

Bugfixes

• fixed backwards incompatible utils changes

[1.0.5] - 2019-06-03

Bugfixes

• fixed spacy being a required dependency (regression)

[1.0.4] - 2019-06-03

Features

• automatic creation of index on the sender_id column when using an SQL tracker store. If you have an existing data and you are running into performance issues, please make sure to add an index manually using CREATE INDEX event_idx_sender_id ON events (sender_id);.

Improvements

• NLU evaluation in cross-validation mode now also provides intent/entity reports, confusion matrix, etc.

[1.0.3] - 2019-05-30

Bugfixes

• non-ascii characters render correctly in stories generated from interactive learning

• validate domain file before usage, e.g. print proper error messages if domain file is invalid instead of raising errors

[1.0.2] - 2019-05-29

Features

• added domain_warnings() method to Domain which returns a dict containing the diff between supplied actions, intents, entities, slots and what's contained in the domain

Bugfixes

• fix lookup table files failed to load issues/3622

• buttons can now be properly selected during cmdline chat or when in interactive learning

• set slots correctly when events are added through the API

• mapping policy no longer ignores NLU threshold

• mapping policy priority is correctly persisted

[1.0.1] - 2019-05-21

Bugfixes

• updated installation command in docs for Rasa X

[1.0.0] - 2019-05-21

Features

• added arguments to set the file paths for interactive training

• added quick reply representation for command-line output

• added option to specify custom button type for Facebook buttons

• added tracker store persisting trackers into a SQL database (SQLTrackerStore)

• added rasa command line interface and API

• Rasa HTTP training endpoint at POST /jobs. This endpoint will train a combined Rasa Core and NLU model

• ReminderCancelled(action_name) event to cancel given action_name reminder for current user

• Rasa HTTP intent evaluation endpoint at POST /intentEvaluation. This endpoints performs an intent evaluation of a Rasa model

• option to create template for new utterance action in interactive learning

• you can now choose actions previously created in the same session in interactive learning

• add formatter 'black'

• channel-specific utterances via the - "channel": key in utterance templates

• arbitrary json messages via the - "custom": key in utterance templates and via utter_custom_json() method in custom actions

• support to load sub skills (domain, stories, nlu data)

• support to select which sub skills to load through import section in config.yml

• support for spaCy 2.1

• a model for an agent can now also be loaded from a remote storage

• log level can be set via environment variable LOG_LEVEL

• add --store-uncompressed to train command to not compress Rasa model

• log level of libraries, such as tensorflow, can be set via environment variable LOG_LEVEL_LIBRARIES

• if no spaCy model is linked upon building a spaCy pipeline, an appropriate error message is now raised with instructions for linking one

Improvements

• renamed all CLI parameters containing any _ to use dashes - instead (GNU standard)

• renamed rasa_core package to rasa.core

• for interactive learning only include manually annotated and ner_crf entities in nlu export

• made message_id an additional argument to interpreter.parse

• changed removing punctuation logic in WhitespaceTokenizer

• training_processes in the Rasa NLU data router have been renamed to worker_processes

• created a common utils package rasa.utils for nlu and core, common methods like read_yaml moved there

• removed --num_threads from run command (server will be asynchronous but running in a single thread)

• the _check_token() method in RasaChat now authenticates against /auth/verify instead of /user

• removed --pre_load from run command (Rasa NLU server will just have a maximum of one model and that model will be loaded by default)

• changed file format of a stored trained model from the Rasa NLU server to tar.gz

• train command uses fallback config if an invalid config is given

• test command now compares multiple models if a list of model files is provided for the argument --model

• Merged rasa.core and rasa.nlu server into a single server. See swagger file in docs/_static/spec/server.yaml for available endpoints.

• utter_custom_message() method in rasa_core_sdk has been renamed to utter_elements()

• updated dependencies. as part of this, models for spacy need to be reinstalled for 2.1 (from 2.0)

• make sure all command line arguments for rasa test and rasa interactive are actually used, removed arguments that were not used at all (e.g. --core for rasa test)

Deprecations and Removals

• removed possibility to execute python -m rasa_core.train etc. (e.g. scripts in rasa.core and rasa.nlu). Use the CLI for rasa instead, e.g. rasa train core.

• removed _sklearn_numpy_warning_fix from the SklearnIntentClassifier

• removed Dispatcher class from core

• removed projects: the Rasa NLU server now has a maximum of one model at a time loaded.

Bugfixes

• evaluating core stories with two stage fallback gave an error, trying to handle None for a policy

• the /evaluate route for the Rasa NLU server now runs evaluation in a parallel process, which prevents the currently loaded model unloading

• added missing implementation of the keys() function for the Redis Tracker Store

• in interactive learning: only updates entity values if user changes annotation

• log options from the command line interface are applied (they overwrite the environment variable)

• all message arguments (kwargs in dispatcher.utter methods, as well as template args) are now sent through to output channels

• utterance templates defined in actions are checked for existence upon training a new agent, and a warning is thrown before training if one is missing