Notes RAG

A smart Q&A system for my personal notes using vector search and AI.

Note: This is based on my personal notes, therefore code changes would be needed for others to use it.

What it does

Ask questions about my notes and get intelligent answers with source links.

The system automatically chunks my markdown notes, stores them in a vector database (Pinecone), and enables the use of AI to provide contextual responses.

Key features:

• Automatic note indexing with CI/CD
• Smart chunking of markdown sections and lists
• Vector-based semantic search
• AI-powered question answering
• Source file tracking and incremental updates
• Evaluation of AI-generated answers quality

Workflow

1. Index your notes - Run the indexer to chunk and store your markdown files in the vector database
2. Ask questions - Choose one of two approaches:
   • Simple Q&A: Use the ask command to generate AI prompts with relevant context
   • Advanced queries: Use agentic AI tools (like opencode or claude code) that connect directly (e.g mcp) to the vector database for complex, multi-step questions

Setup

Install dependencies:

Note: Use uv run nvim to launch your editor with the correct virtual environment.

uv sync

Create .env file with:

PINECONE_API_KEY=your_api_key
OLLAMA_HOST=http://localhost:11434  # optional for local AI question enhancement

Usage

Quick commands:

# Index your notes (*-test for testing vector db)
mise run indexer-prod

# Ask a question and get the prompt in the clipboard
mise run ask "what did i do the last week?"

Manual usage:

See mise.toml or scripts in src/ for additional commands.

# Index notes (defaults to ~/Documents/notes)
uv run src/ai_notes_indexer.py --prod --root /path/to/notes

Testing

# Run unit tests
mise run test

# Evaluate answer quality
mise run evaluate