AI Personal Knowledge Base (Second Brain)

An end-to-end RAG-based personal knowledge base. Upload PDFs, notes, bookmarks, and research papers. Chat naturally to query your documents. A critic agent validates responses for factual consistency.

Features

• Upload and manage documents (PDF, text, markdown, HTML)
• Ingestion: parse, chunk, and embed content
• Vector search + RAG generation
• Critic agent that checks answers against sources
• Minimal web UI (uploads + chat)
• Dockerized deployment

Quickstart (Local)

1. Copy .env.example to .env and update keys
2. Install Docker and Docker Compose
3. Run:

docker compose up --build

4. Open the UI at http://localhost:5173
5. API docs at http://localhost:8000/docs

Development (no Docker)

• Backend (Python 3.11+):

python -m venv .venv
. .venv/Scripts/activate  # Windows PowerShell: .venv\Scripts\Activate.ps1
pip install -r backend/requirements.txt
uvicorn app.main:app --app-dir backend --reload --host 0.0.0.0 --port 8000

• Frontend:

cd frontend
npm i
npm run dev -- --host

Configuration

Set environment variables in .env:

• OPENAI_API_BASE, OPENAI_API_KEY: OpenAI-compatible API
• OPENAI_MODEL, OPENAI_EMBED_MODEL: chat and embedding models
• DATA_DIR, VECTOR_DB_DIR, UPLOAD_DIR: storage paths
• CORS_ALLOW_ORIGINS: comma-separated or *

Stack

• Backend: FastAPI, LangChain, FAISS
• LLM: OpenAI-compatible API
• Frontend: React + TypeScript

Deployment

• docker-compose.yml runs backend and frontend services
• Bind mounts storage/ for persistence

Notes

• For air-gapped or self-hosted LLMs, set OPENAI_API_BASE to your server
• PDFs parsed with pypdf; HTML/Markdown with BeautifulSoup and markdown-it-py