self-notes

Local RAG system for querying Notion workspaces. Runs entirely on your machine using Ollama.

Requirements

• Python 3.10+
• Ollama

Setup

Ollama Service management:

brew services start ollama    # Start
brew services stop ollama     # Stop
brew services restart ollama  # Restart
brew services list            # Check status

Ollama Pull Models

ollama pull mistral
ollama pull nomic-embed-text

Python

# Create and activate venv
python3 -m venv .venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Configure
cp .env.example .env
# Add NOTION_API_KEY to .env

Notion Integration

1. Create integration at https://notion.so/my-integrations
2. Copy the internal integration token to .env
3. Share target pages with the integration (page menu → Connections → add integration)
4. Add page URLs to ROOT_PAGES in src/config.py

Usage

CLI

python cli.py index                     # Sync and index from Notion
python cli.py ask "your question"       # Query indexed content
python cli.py sources                   # List indexed pages

API Server (for Open WebUI)

Start the OpenAI-compatible API server:

python api.py

The server runs at http://localhost:8000.

Run Open WebUI in Docker:

docker run -d \
  -p 3000:8080 \
  -v open-webui:/app/backend/data \
  --name open-webui \
  --add-host=host.docker.internal:host-gateway \
  ghcr.io/open-webui/open-webui:main

Open WebUI will be available at http://localhost:3000.

Connect to self-notes:

1. Setup Admin Account (if first time)
2. Go to Admin Panel → Settings → Connections
3. Add a new OpenAI-compatible connection:
   • URL: http://host.docker.internal:8000/v1
   • API Key: any-value (not validated)
4. Select the self-notes model in your chat

Architecture

Notion API → fetch pages (recursive) → chunk → embed (Ollama) → numpy vector store
                                                                       ↓
Question → embed → cosine similarity search → context + prompt → Ollama → answer

Configuration

src/config.py:

Variable	Description
ROOT_PAGES	List of Notion page URLs/IDs to index (includes children recursively)
EMBEDDING_MODEL	Ollama model for embeddings (default: nomic-embed-text)
LLM_MODEL	Ollama model for generation (default: mistral)
CHUNK_SIZE	Text chunk size in chars (default: 500)
CHUNK_OVERLAP	Overlap between chunks (default: 50)
TOP_K_RESULTS	Number of chunks to retrieve per query (default: 4)

Project Structure

src/
├── config.py       # Configuration
├── notion.py       # Notion API client, recursive page fetching
├── indexer.py      # Chunking, embedding, storage
├── query.py        # Retrieval and LLM generation
└── vectorstore.py  # Simple numpy-based vector store
cli.py              # CLI entrypoint
db/                 # Vector store data (gitignored)

Notes

• Only pages shared with your integration are accessible
• Child pages are fetched recursively from each root page
• Re-running index clears and rebuilds the entire index
• All data stays local except Notion API calls to fetch content
• Uses numpy for vector similarity - no external DB dependencies