Skip to content
Scan Phase 3

Scan a shelf

Photograph a shelf. A vision model reads the spines and drafts catalogue rows; nothing is saved until you confirm each one, and the photo itself is never stored.

Scanning is off on this catalogueReading spines needs a vision model, and a vision model costs money per photo. This deployment runs with VISION_ENABLED=false, so no photo is sent anywhere and nothing is charged. It is switched on only on the owner's own instance, behind a hard monthly cap.
When it runs
What a scan does
  • Reads up to 60 spines per photo and drafts one row for each: the text, a bounding box, a colour.
  • Resolves each draft against the catalogue by title and author. Unreadable spines stay unresolved rather than guessed.
  • Uses a credit card in frame, if there is one, to put millimetres on the widths.
  • Confirmed rows become shelf_crop claims at the model's confidence × 0.9 — yours, but not measurements. They resolve as Estimated, never Verified.
Privacy

A shelf photo is a picture of somebody's home. It is never persisted, records default to hashes only with empty tags, and no face work is done at any point — people in frame are ignored.