RAG Demo Chat (Socket.IO Streaming)

Production-ready, Dockerized, full-stack LLM chat app.

Prerequisites

1. Docker + docker-compose
2. Set your OpenAI API key in .env (required for chats to work):
   • OPENAI_API_KEY=...
   • Optional: OPENAI_MODEL (defaults to gpt-4o-mini)
3. Vector memory uses Qdrant + MongoDB (provided by docker-compose).

If you want to override defaults, you can also set:

• MONGODB_URI, MONGODB_DB
• QDRANT_URL
• OPENAI_EMBEDDING_MODEL, EMBEDDING_DIMENSION

Run

docker-compose up --build

After containers are up:

• Frontend: http://localhost:3000
• Backend: Socket.IO endpoint at http://localhost:4000/ws (Socket.IO path /ws)

The frontend connects to that Socket.IO endpoint directly; Docker is configured so localhost:4000 routes to the backend container.

Architecture (high level)

• frontend/ (Next.js App Router, no Next API routes) connects to the backend via Socket.IO.
• backend/ (Express + socket.io) calls OpenAI with streaming enabled and forwards streamed tokens over Socket.IO events.