Side Quest: an AI that sends you on a walk, and never learns where you are

Description and use

Side Quest is a walking scavenger hunt that generates clues in the browser using an open‑source LLM, keeps the user’s exact location private, and relies on open data sources.

Problem

Most AI‑outdoors apps keep users looking at the screen while walking, reducing real‑world engagement.

Outcome

Side Quest delivers a privacy‑preserving, low‑cost walking scavenger hunt that limits screen time to ~10 s per stop, keeps location data local, works offline after first run, and uses open data sources.

Workflow

• 1. User selects walk duration (15–60 min) and clue style (playful, poetic, detective, kids).
• 2. App snaps current location to a 0.01° (~1.1 km) grid and queries OpenStreetMap and Wikipedia for nearby points of interest.
• 3. A deterministic planner selects 2–4 varied stops that fit the requested time.
• 4. Gemma (WebLLM on WebGPU) writes a riddle and a sensory notice for each stop.
• 5. User walks, checks in at each stop, receives the notice, and proceeds to the next clue.
• 6. After the final stop, Gemma writes a three‑line postcard summarizing the walk, saved only on the device.

Evidence

Gemma writes a riddle for each stop

Source

The app finds a few interesting places within walking distance (a temple, a mural, an old bookshop, a pond) and plans a loop that fits your time.

Source

Gemma, running inside your browser, writes a riddle for each stop.

Source

The only network requests are the map lookups, and even those never carry your exact position.

Source