What it does

Rasa is a platform for product teams building customer-facing AI agents. Its approach, called CALM, pairs the flexibility of large language models with structured flows, deterministic logic and built-in recovery patterns, so a team can let the model drive in some places and add order and conditions in others. Teams start from a prompt and tune it in a visual workspace, in code or with a coding agent. Rasa then runs the agent through simulated conversations in voice and text and shows what passed, supports end-to-end checks and version comparison, and reviews live conversations to point out what to fix. The platform covers chat, voice, retrieval of answers from company data, intent-based NLU, multilingual agents, orchestration of agents and tools, and MCP for connecting agents to APIs as tools. The site says agents can run on your own infrastructure with your choice of model, and can serve channels such as a hotline, Slack, web chat, WhatsApp and email. A free Developer Edition gives access to Rasa with CALM. Other pricing is not given on the page.

Capabilities

CALM approach that combines LLM flexibility with deterministic flows and recovery patterns
Automatic evaluation of agents on simulated voice and text conversations
Studio workspace for visual building, review and monitoring, with changes saved as code
CLI and SDKs for building, testing and debugging in a version-controlled codebase
Voice, chat, multilingual and multi-channel agents
Enterprise RAG, orchestration and MCP tool connections

When to use it

Building customer support or sales assistants that need predictable, explainable behavior
Running voice agents with audit trails and secure deployment
Deploying agents on your own infrastructure with your own choice of model
Testing agent changes against simulated conversations before taking them live