Rasa
Industrial-grade open-source conversational AI framework, now with CALM
- License
- Apache-2.0 (OSS) / Pro commercial
- Deploy difficulty
- Hard
- GitHub stars
- 21.2k
- Last commit
- 109d ago
Latest
3.6.21 4.9k forks
145 open issues Synced 2026-05-17 What Rasa is#
Rasa is the longest-standing open-source conversational AI framework. The 2024 CALM paradigm uses LLMs for understanding and declarative flows for control, resolving the old tension between uncontrollable LLMs and chat-blind state machines.
When to choose Rasa#
Reach for it when:
- Compliance requires data to stay on-prem
- You already have ML / Python engineers
- Logic is complex and needs strict state machines with LLM fallback
Most SMB teams should start with Dify or Chatwoot.
Strengths
- New CALM paradigm decouples flow logic from LLMs
- Fully local inference — compliance friendly
- Declarative flows + LLM NLU
- Years of industrial deployment experience
Trade-offs
- Highest learning curve — needs Python engineering
- Training data, action server and flow debugging are heavy
- Clear split between OSS and Pro features
Best for
- Regulated industries — finance, government, healthcare
- Teams with Python / ML engineers needing fine-grained control