🌿 Trail Kit
Offline-first, open-source AI tools that get you off the screen and into the world.
Built for the DEV “Touch Grass” challenge. Everything runs locally with open-weight models, so it works on a trail with no signal, keeps your data on your device, and costs nothing to run.
- 🐦 Trail Birder: record a bird call and identify the species on-device, even in airplane mode.
- 🌱 Frost Gardener: enter your local frost dates and see what to plant this week. A local LLM explains why.
- 🍁 Foliage Run Club: build group-run routes from open map data, ranked by tree cover and peak colour.
Why open source?
- No signal needed: open weights run where cloud APIs can’t.
- Private: your location and recordings never leave your device.
- Swappable: fine-tune or swap models without rewriting the app.
- Free: no per-call fees, so run clubs and school gardens can use it.
Quick start
git clone https://github.com/YOUR-USERNAME/trail-kit
cd trail-kit
ollama pull llama3.2
python -m trailkit serve
To try the garden planner demo only, open index.html in your browser.
Tech stack
- Ollama / llama.cpp for local inference
- An open bird-sound classification model
- OpenStreetMap data
- SQLite
Contributing (Hacktoberfest welcome! 🎃)
Contributions of all sizes are welcome. Good first issues:
- Add frost-date presets for your region
- Add crops to the planner table
- Translate the UI into your language
- Benchmark a new open model for bird calls
Look for issues labelled good first issue and hacktoberfest. See CONTRIBUTING.md for how to open a pull request.
License
MIT. See LICENSE.