Behind the Release Notes: Rasa 3.6

Posted Jun 30, 2023

Updated

With Rasa 3.6, conversational AI teams get powerful new tools to lock down sensitive data and boost AI Assistant performance at scale. Key updates include advanced PII Data Management and Load Testing Guidelines designed to help your AI Assistants run securely and smoothly in demanding enterprise environments.

PII Data Management

In conversational AI, safeguarding personally identifiable information (PII) is non-negotiable. Rasa 3.6 introduces a multi-layered PII detection and anonymization system that keeps your AI assistants compliant and trustworthy without sacrificing usability.

How Rasa Finds PII:

Slot-Based Detection: Leverage your domain knowledge by defining sensitive slots in your conversation flows. Rasa automatically flags PII during slot filling, catching known sensitive data points reliably.
GLiNER PII Model: For unexpected or free-text PII, integrate the optional GLiNER model to detect hard-to-catch entities missed by traditional slot methods.

Flexible Anonymization Options:

Redaction: Replace sensitive values with characters (e.g., 1234-5678-****-3456 for credit cards)
Masking: Substitute with semantic labels (e.g., [EMAIL_ADDRESS])

You control what remains visible, balancing privacy protection and user experience.

Load Testing Guidelines

We now introduced load testing guidelines for enterprise AI Assistants at scale to help better estimate the resources required for operation in production. We have also provided some guidance detailing how to run load-tests on your Rasa Assistants.

Additional Enhancements

Other Important Updates in Rasa 3.6

End of Python 3.7 Support: Python 3.7 reached end-of-life in June 2023. Upgrade to supported versions to keep your environment secure and efficient.
ARM Architecture Support: Run Rasa Pro smoothly on arm64 processors, including Apple M1/M2 chips (available in Rasa 3.6.2+).

Explore all updates and bug fixes in the full changelog here.