This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass
What I Built
Trail Kit is an offline-first, open-source toolkit that helps people get outside:
- 🐦 Trail Birder: identify bird calls on-device, with no signal. Status: in progress
- 🌱 Frost Gardener: enter your local frost dates and see what to plant this week. Status: built (garden planner is live)
- 🍁 Foliage Run Club: build group-run routes from open map data. Status: planned
The screen is meant to be the shortest part of the experience: check what to plant, then go plant it. Hear a bird, then look up.
It's for hikers, birders, gardeners and run clubs who want technology that stays out of the way, and for school gardens and clubs that can't pay per-call API fees.
Demo
- 🌐 Live site: https://tanishq964.github.io/trail-kit./
- 💻 Code: https://github.com/Tanishq964/trail-kit./blob/main/Trail%20Kit%20%E2%80%94%20open-source%20AI%20that%20gets%20you%20outside.html
The live page has a working garden planner that runs in your browser. The local-model parts (the LLM explanation and bird-call identification) run from the repo on your own machine with Ollama. They are not yet wired into the hosted web page.
Code
Full source, MIT licensed: https://github.com/Tanishq964/trail-kit.
How I Built It
Trail Kit is built around open pieces:
- Model: Llama 3.2, an open-weight model, explains the planting advice in plain language. An open bird-sound classification model handles bird-call identification.
- Inference: Ollama (llama.cpp underneath), running locally.
- Data: OpenStreetMap for routes and tree cover, plus user-entered frost dates.
- Storage: SQLite, all on-device.
- License: MIT.
The garden planner takes your last and first frost dates, compares them with each crop's planting window, and lists what to plant this week. The local LLM then explains why, so the advice is readable instead of just a table.
Why Does Open Innovation Matter?
- It works where the trail has no signal. A cloud API can't identify a bird when you have zero bars. Open weights running locally can.
- Your data stays yours. Your location and recordings never leave your device.
- I can swap models. Because inference goes through Ollama, changing the model is a one-line change, with no app rewrite.
- It costs nothing to run. No per-call fees means a run club or school garden can use it freely.
For an outdoor tool, this is the deciding difference: a closed API fails exactly where this app is meant to be used.
Honest Status
- I have not yet field-tested Trail Birder on a real trail, so I'm not claiming accuracy numbers.
- The bird model and run-route builder are not yet connected to the live page.
- Taking it outside and reporting back is the next step.
Open for Contributions
Trail Kit welcomes contributors this Hacktoberfest. Beginner-friendly issues include:
- Adding frost-date presets for more regions
- Adding more crops to the planner
- Translating the UI
- A light/dark theme toggle
- Benchmarking open bird-sound models
Issues: https://github.com/Tanishq964/trail-kit./issues
What's Next
- Wire the bird model into the web page
- Build the foliage route ranking from OpenStreetMap tree cover
- Add offline map tiles so routes work with no signal
Top comments (0)