title	Anubodh AI
emoji	🧠
colorFrom	indigo
colorTo	purple
sdk	docker
app_port	7860
pinned	false

Anubodh AI - Personalized RAG LLM ChatBot

Anubodh AI is a production-grade, multi-user, context-adaptive personal AI assistant platform built with a FastAPI backend and a Next.js frontend. The system features continuous background memory learning, vector database retrieval, and support for the Groq API.

---

Key Features

• Groq LLM Service: Powered by llama-3.3-70b-versatile for fast streaming chat responses and precise structured JSON memory extraction.
• Continuous Background Memory Learning: A background agent parses conversation turns in real-time, extracting preferences, goals, relationships, and facts.
• Robust Hybrid RAG Memory:
  • Retrieves relevant context using Qdrant vector search.
  • If vector embeddings are rate-limited or unavailable, it automatically falls back to fetching recent active profile facts directly from SQLite.
• Response Length Control Toggle: Tester-friendly button selectors on the frontend (Very Short = 1 sentence, Short = 2-3 sentences, Medium = up to 5 sentences).
• Conflict Resolution: The memory agent automatically deactivates outdated or contradicting memories when a user updates their preference.
• Windows Compatibility: Unicode emojis removed from logs to prevent terminal encoding failures.

---

Tech Stack

• Frontend: Next.js, Tailwind CSS, TypeScript, Zustand (State Management)
• Backend: FastAPI, Uvicorn, SQLAlchemy (Async), SQLite
• Vector Database: Qdrant Client (Persistent Local Database / In-Memory)
• LLM Engine: Groq SDK (llama-3.3-70b-versatile)
• Embeddings: Google Gemini API (text-embedding-004) with safe mock vector fallback.

---

Installation & Local Setup

Step 1: Configure Environment Variables

Create a .env file in the root folder (or copy .env.example to .env):

# Groq API Key (Required for chat and memory extraction)
GROQ_API_KEY=your_groq_api_key_here

# Google Gemini API Key (Optional for embeddings fallback)
GOOGLE_API_KEY=your_gemini_api_key_here

DATABASE_URL=sqlite+aiosqlite:///./anubodh.db
QDRANT_URL=memory

Step 2: Start the Backend (FastAPI)

1. Navigate to the backend directory:

   cd backend

2. Create and activate a Python virtual environment:

   # Windows
   python -m venv venv
   .\venv\Scripts\Activate.ps1
   
   # macOS/Linux
   python3 -m venv venv
   source venv/bin/activate

3. Install dependencies:

   pip install -r requirements.txt

4. Run the development server:

   python -m app.main

   The Swagger interactive API documentation will be available at http://localhost:8000/docs.

Step 3: Start the Frontend (Next.js)

1. Open a new terminal and navigate to the frontend directory:

   cd frontend

2. Install Node modules:

   npm install

3. Start the Next.js dev server:

   npm run dev

   The user interface will be available at http://localhost:3000.

---

Running Automated Backend Tests

We include a test runner to verify database integrity, Groq connectivity, memory extraction, and fallbacks.

Navigate to the backend directory and run:

python test_services.py

---

License

Open-source under the MIT License.