RasaGPT glues Rasa to Langchain because the alternative was nothing

What it does

RasaGPT is a headless chatbot boilerplate that straps a Langchain and LlamaIndex retrieval pipeline onto Rasa, exposing everything through a FastAPI backend and a Telegram frontend. It handles document uploads, versioning, and “re-training” by chunking your corpus into pgvector-backed Postgres, then injecting relevant context into conversations. The author built it after scouring GitHub and finding zero decent reference implementations of LLMs integrated with Rasa.

The interesting bit

The real value isn’t the bot; it’s the glue. The project sorts out tedious but critical wiring that Rasa doesn’t natively offer: multi-tenancy, session metadata passing, a custom pgvector schema that sidesteps Langchain’s opinionated PGVector class, and even a Docker workaround for Rasa on Apple Silicon. Think of it as an integration cookbook with a working demo attached.

Key highlights

• Document-aware conversations via Langchain/LlamaIndex retrieval and pgvector similarity search
• Multi-tenant schema and session handling that Rasa itself lacks
• FastAPI backend with auto-generated Swagger docs and async endpoints
• Human handoff detection and automatic conversation tagging
• Bundled ngrok tunneling and Telegram webhook setup

Caveats

• The author explicitly warns this is “far from production code” and rife with prompt injection and security vulnerabilities
• Retrieval quality is currently “hit or miss” using basic 1000-character chunking and euclidean distance; the author estimates optimization could improve output quality by more than 70%, but that work isn’t done
• Non-MacOS users must manually swap Docker images, suggesting macOS was the primary development target

Verdict

Grab this if you need a working reference for bolting modern LLM retrieval onto an existing Rasa deployment. Skip it if you want a secure, production-ready chatbot platform out of the box.