Chatbot with Memory & Context (RAG)

This project demonstrates a production-ready chatbot backend using:

• Node.js + TypeScript
• MongoDB (conversation + long-term memory)
• MongoDB Atlas Vector Search (semantic memory)
• OpenAI LLM

Features

• Conversational memory
• Semantic recall using vector search
• Long-term fact extraction
• Clean architecture (API / Application / Domain / Infra)
• Easily extensible for agents, tools, RBAC

MongoDB Vector Index Setup

Before running, create the memory_index on your memories collection with the following mapping:

{
  "fields": [
    {
      "numDimensions": 768,
      "path": "embedding",
      "similarity": "cosine",
      "type": "vector"
    },
    {
      "path": "userId",
      "type": "filter"
    },
    {
      "path": "importance",
      "type": "filter"
    }
  ]
}

Refer to MongoDB Atlas documentation for details on creating vector indexes.

Run

npm install
cp .env.example .env
npm run dev

API

POST /chat

{
  "userId": "user1",
  "message": "My name is Yogesh"
}