Built for the Touch Grass week of Hacktoberfest
Most apps are designed to keep you in. Groundline is designed to get you out.
You open it, tell it two things (how long you have and what kind of outing you want), and it hands you one pocket-sized card:
đ A destination, a real park near you
đ A sensory cue, something to notice when you get there
âąď¸ A timer, then the phone goes back in your pocket
No feed, no infinite scroll. You do one quick check, then leave.
How it works
Before the model writes a single word, Groundline gathers real local data:
- OpenStreetMap supplies verified nearby parks and green spaces, so it never sends you somewhere that doesnât exist.
- Open-Meteo supplies live weather, so a 20-minute walk is planned around the rain thatâs about to arrive.
- Backboard routes the request to a model and keeps assistant memory, so the app knows which places youâve already visited and can nudge you toward somewhere new.
- An open-weight model (Gemma, Llama or Qwen) writes the card from that grounded prompt.
The model doesnât invent the destination. It only turns facts into a short, human-sounding card.
Why Backboard is central
Two things make the card feel personal instead of generic:
- It remembers. Places youâve already been get deprioritized, so the app keeps nudging you somewhere new.
- Itâs grounded. The final prompt is built from real OSM and weather data, which keeps the model from hallucinating a lake.
Backboard also makes swapping between open-weight models easy, so you can try Gemma, Llama or Qwen and see how each one changes the voice of the card.
No model? No problem
If the model is unavailable, a fallback almanac path still produces a card from the same local data. The core promise (a destination, a cue and a timer) survives without AI.
Groundline is lightweight and meant to be forked. Swap in your cityâs data, change the tone of the cards, or plug in a different model.
đ Github Repo: Groundline
If you take a walk because of it, thatâs the whole point. đł





Top comments (0)