Behind the Release Notes: Rasa 3.7

Posted Nov 29, 2023

The Rasa Platform is an enterprise-grade generative conversational AI platform designed for building, deploying, scaling, securing, and monitoring AI agents. The platform combines the pro-code flexibility of the Rasa framework with the no-code speed of Rasa Studio enabling entire teams to collaborate efficiently across the conversational AI lifecycle.

• Who is it for? Rasa is built for enterprise CAI teams, including developers, data scientists, ML engineers, test managers, solution architects, conversation designers, flow builders, and content teams. It lets these teams move fast without compromising on reliability, control, or scale.

What’s New in Rasa 3.7:

1. Conversational AI with Language Models (CALM)

CALM is our LLM-native approach to building agents that are more fluent, contextual, and resilient-without sacrificing control. CALM isn’t just a feature set-it’s a new conversational architecture powered by language models at the core.

Highlights:

• Intentless understanding: CALM agents don't rely on intent classification to understand user input. Instead, they interpret messages directly in context.
• Out-of-the-box patterns: Rasa agents handle digressions, interruptions, and corrections naturally-without extra logic.
• Contextual responses: Optionally rephrase templated responses to sound more fluid based on context.
• Hard guardrails: CALM agents only respond using approved templates. Generated responses are opt-in and tightly scoped.

How it works:

• User message received: The Dialogue Understanding module parses input in the full context of the conversation.
• Action plan created: A language model generates a 'command'-a proposed plan for what should happen next.
• Dialogue Manager executes: The command is evaluated by available policies to determine the next step
  • A Flow is activated, switched, or canceled. Flows handle dialogue and custom actions.
  • (Optional) A search is triggered to a connected vector database.
  • (Optional) A templated response is rephrased before delivery.

Want a visual walkthrough? Follow along in this video.

2. Flows

Flows are modular, pre-defined user journeys enabled by the new FlowPolicy. With Flows, you can encode business logic in code or in our no-code UI Rasa Studio.

• With Rasa, you can optionally allow the dialogue system to handle flow activation at times, and also continue using NLU where you see fit to activate flows.
• Integrate custom actions directly into flows.

3. Enterprise Search

The Enterprise Search Policy lets agents search your internal knowledge base in real time. Built with RAG (retrieval augmented generation), it integrates directly with your vector DB of choice.

• Search is a first-class policy-CALM can choose between flows and search.
• Use it with CALM or an NLU-only agent.
• Programmatically activate search from within a flow if needed.

4. Contextual Response Rephraser

The contextual response rephraser improves templated replies by adjusting tone, grammar, and clarity based on user context.

• Toggle rephrasing on or off per response.
• Useful for reducing repetition, boosting clarity, and personalizing tone.

See it in action: Watch the video by ML Engineer Tanja Bunk.

Want to learn more about Rasa and how you can get access to these new capabilities? Connect with the team or try it out yourself.