Spotter

Spotter is a workout form coach built for one person: a friend who trains at home, with no gym nearby and no budget for a private trainer. You film a short clip of a set, and Spotter counts your reps, points out form issues rep by rep, and writes a short coaching note with fixes and a plan for next time — plus a video with the issues marked on it.

What makes it different is that your video never has to leave your own computer. Small open models do the work on your machine: one reads your body's positions from the video, a tiny trained model works out which exercise it is, and plain code counts the reps before any coaching words are written. A checker then blocks any claim the numbers do not back up, so the coach says less rather than inventing advice.

Vibe-coded evidence, in the author's own words on DEV (5 October 2026): "I built this with Claude Code in a long session of small, verified steps." The repository's own tools list also names Claude Code and OpenAI Codex as repo-aware coding agents. GitHub stars: 0, checked on 29 October 2026.

Honest limits: Spotter is not a medical device — it does not diagnose injuries or replace a trainer. It only knows three moves (squat, push-up, shoulder press) and deliberately says "I can't tell" for anything else instead of guessing. And there is no working hosted demo right now: the Hugging Face Space currently shows only a static placeholder page and the Render link returns 404, so the demo link is the repository page and the app runs locally (uv sync, then uv run python app.py) — the description says so plainly.