hunchml

Local AutoML for tabular data. Bring a spreadsheet, pick the column you want to predict, and get a trained, explained model, without writing code. Everything runs on your computer; your data never leaves it.

Use it

pip install hunchml
hunchml

On a Mac, also run brew install libomp once (Homebrew): two of the model types (LightGBM and XGBoost) need it, and without it the app skips them.

Your browser opens the app. Keep the terminal window open while you use it; press Ctrl+C there to stop. Options: --data-dir (where projects are stored, default ~/.hunchml), --port, --max-concurrent (training runs at once, default 1) and --no-browser.

Develop

Needs uv and Node.js 24.

uv sync                                  # Python environment
(cd frontend && npm ci)                  # frontend dependencies

uv run pytest                            # fast Python tests (slow ones: -m slow)
uv run ruff check src tests scripts && uv run mypy
(cd frontend && npm test && npm run lint && npm run typecheck)

Work on the interface with live reload: start the backend with uv run hunchml --no-browser, run npm run dev in frontend/, and open the address Vite prints with the ?token=... part of the backend's link added (for example http://localhost:5173/?token=abc). Vite forwards /api calls to the backend on port 8765.

npm run build in frontend/ writes the interface into src/hunchml/server/static/, which the backend serves and the wheel ships. The end-to-end test needs that build: npm run build && npx playwright install chromium && npm run e2e.

Release

Set the version in pyproject.toml, commit, and push a tag v<version>. The Release workflow builds the interface and the package, checks both, and publishes to PyPI.