NotionHealth AI is a local-first health tracker for Notion. You run it on your own machine, connect it to Claude through MCP if you want the assistant workflow, and use the same local setup for the browser UI.

The project covers day-to-day health logging, summaries, goals, symptoms, appointments, AI-written insights, and optional TRIBEv2 brain analysis. If TRIBEv2 is not installed, the rest of the app still works and the brain features fall back to simulation mode.

⚠️ Consult a qualified health professional for medical advice, diagnosis, or treatment.

git clone https://github.com/rotsl/notion-health-ai.git
cd notion-health-ai
python3 -m venv venv
source venv/bin/activate
pip install -e .
cp .env.example .env

On Apple Silicon, use a native arm64 Python if you want TRIBEv2:

/opt/homebrew/bin/python3.11 -m venv venv_arm64
source venv_arm64/bin/activate
pip install -e .

Set at least these values in .env:

NOTION_API_KEY=your_notion_token_here
ANTHROPIC_API_KEY=your_anthropic_key_here

Add this if you want the real TRIBEv2 model path:

HUGGING_FACE_TOKEN=hf_your_token_here

Share a Notion page with your integration, then run:

python scripts/setup_notion.py

The script creates the databases, seeds sample data, builds the dashboard, and writes the generated database IDs back into .env.

source venv/bin/activate
make serve-api

Open http://127.0.0.1:8000.

If you prefer the direct command:

uvicorn notion_health_ai.api:app --reload --host 127.0.0.1 --port 8000

To stop the local app:

Ctrl+C

Useful local checks:

curl http://127.0.0.1:8000/api/status
pytest tests/ -q

If port 8000 is still busy after stopping the app, find the process first:

lsof -nP -iTCP:8000 -sTCP:LISTEN

Then stop it cleanly:

kill <PID>

Only if the process is stuck, force it:

kill -9 <PID>

The browser UI runs from the same local FastAPI server. Open the Settings panel and provide:

The browser stores those values in session storage only.

The main tabs:

• Brain Predict: text, audio, video, and multimodal TRIBEv2 analysis
• Health Log: weight, sleep, mood, energy, steps, water, exercise, blood pressure, heart rate, and notes
• AI Insights: generated summaries from your Notion health data
• Visualizations: generated brain outputs saved under visualizations/

The Gallery now has a Clear Saved Outputs button. It deletes generated visualization files from the local visualizations/ directory and keeps the placeholder file.

TRIBEv2 is optional. If you want the real model path:

1. Request access to meta-llama/Llama-3.2-3B on Hugging Face
2. Create a read token
3. Put that token in HUGGING_FACE_TOKEN
4. Install the heavier local dependencies in the same environment you use to run the app

Without those pieces, brain features fall back to simulation mode.

For text-based brain analysis in the browser UI, the app asks for a simulation duration before it runs. If your connected Notion data already contains useful timing data, the app can offer those values directly. Otherwise it offers a small set of realistic presets and bounds any custom value to a sensible range.

Visualization timelines now follow the selected simulation duration, and generated titles are written to describe the actual simulation instead of a generic label.

If you want the MCP workflow, add the local server command to Claude Desktop:

{
  "mcpServers": {
    "notion-health-ai": {
      "command": "/absolute/path/to/notion-health-ai/venv/bin/notion-health-mcp"
    }
  }
}

If you are using the arm64 environment, point Claude at venv_arm64/bin/notion-health-mcp instead.

Then fully quit and reopen Claude Desktop.

Example prompts for Claude:

Log my weight as 172 pounds, mood 7, and 7.5 hours of sleep

I ran for 40 minutes today and took 9,500 steps

Create a goal to sleep 8 hours a night by the end of the month

Log a headache at severity 6 with light sensitivity

Predict my brain response to 30 minutes of meditation

Give me a cognitive health report for this week

• CONTRIBUTING.md
• SECURITY.md
• CODE_OF_CONDUCT.md
• CHANGELOG.md

The code in this repo is MIT licensed. See LICENSE.

TRIBEv2 is separate from this repo. If you enable the real model path, you also need to follow the upstream Meta and Hugging Face terms for the TRIBEv2 model and the gated meta-llama/Llama-3.2-3B dependency.

At the time this README was updated, the upstream TRIBEv2 pages were not perfectly consistent about the exact model license label, so check the current upstream pages before redistribution or commercial use:

• TRIBEv2 model page: https://huggingface.co/facebook/tribev2
• TRIBEv2 source repo: https://github.com/facebookresearch/tribev2
• Llama 3.2 model page: https://huggingface.co/meta-llama/Llama-3.2-3B