Integrating External Agents via A2A

New Beta Feature in 3.14

Rasa supports stateful execution of external agents via A2A protocol.

This feature is in a beta (experimental) stage and may change in future Rasa versions. We welcome your feedback on this feature.

Overview

Rasa can act as an intelligent orchestrator that coordinates a network of multiple agents through the Agent-to-Agent (A2A) protocol. This is particularly useful when at least one of the agents is externally built, i.e. by a different department, by a third party, or it's a legacy project on a different tech stack. Integrating each of these agents into a single system enables contextually rich conversations through single unified conversational interface.

The Orchestrator Model

In this architecture, your Rasa agent serves as the orchestrator that:

1. Owns the conversation flow - Determines when and which external agents to involve
2. Manages business logic - Enforces business rules, compliance, and conversation patterns
3. Controls agent handoffs - Decides when to delegate tasks and when to resume control
4. Maintains conversation context - Preserves user context across different agent interactions
5. Allows context engineering - Allows fine-grained control over context passed between agents
6. Handles final responses - Ensures consistent user experience regardless of which agent provided the information

Configuring External Sub-Agents

Registering the sub agent

Configure external sub agents in the sub_agents/ folder of your Rasa agent:

your_project/
├── config.yml
├── domain/
├── data/flows/
└── sub_agents/
    ├── analytics_agent/
       └── config.yml

The A2A protocol prescribes using an agent card to expose any externally running agent's capabilities. The agent card can be provided as part of the config.yml defined for each external agent -

sub_agents/analytics_agent/config.yml

# Basic agent information
agent:
  name: analytics_agent
  protocol: a2a
  description: "External agent that provides analytical capabilities"

# A2A configuration
configuration:
  agent_card: ./sub_agents/shopping_agent/agent_card.json

More configuration options can be found in the reference page.

Invoking the External Agent

The orchestrating layer in Rasa provides you the control to invoke an external agent when your business logic needs you to do so by using the call step in a flow -

flows.yml

flows:
  product_enquiry:
    description: Handle questions related to analysis needed on the product
    steps:
    - collect: query
    - call: analytics_agent  # External A2A agent

For more information on how Rasa passes control and receives response from an external agent via A2A, head over to the documentation on External Sub Agents.

Context Management

The orchestrating agent in Rasa can control what context is shared with external agents by customizing the client interfacing with the external agent.

To do so, define a custom python module -

sub_agents/analytics_agent/custom_agent.py

from rasa.agents.protocol.a2a.a2a_agent import A2AAgent
from rasa.agents.schemas import AgentInput

class AnalyticsAgent(A2AAgent):
    async def process_input(self, input: AgentInput) -> AgentInput:
        # Filter slots to only include relevant context
        filtered_slots = [
            slot for slot in input.slots
            if slot.name in ["user_id", "account_type", "inquiry_category"]
        ]

        # Create new input with filtered context
        return AgentInput(
            id=input.id,
            user_message=input.user_message,
            slots=filtered_slots,  # Only share necessary slots
            conversation_history=input.conversation_history,
            events=input.events,
            metadata=input.metadata,
            timestamp=input.timestamp
        )

Provide the customized interface to your sub agent's config:

sub_agents/analytics_agent/config.yml

agent:
  name: analytics_agent
  protocol: a2a
  description: "External agent that provides analytical capabilities"

configuration:
  agent_card: ./sub_agents/analytics_agent/agent_card.json
  module: sub_agents.analytics_agent.custom_agent.AnalyticsAgent

For more information on possible customizations, head over to the reference documentation.