From the bag to your catalog.

1. Photograph the bag

Front and back if the details are split. The photos stay on your device until you send them to your AI.

2. Send to your AI

Copy the instructions and paste them into your AI chat with the photos.

The photos only leave your device when you send or paste them into your AI. They never go through shelf.coffee.

See the full instructions

# shelf.coffee · read a coffee bag Read the attached photos of the bag and reply **only** with JSON in the format below, inside a single json code block, without explanations. ## Errors (reply with the line only) - No photo attached: `{"schemaVersion":1,"error":"missing_photo"}`. Do not use examples or earlier conversations. - Cannot open the images: `{"schemaVersion":1,"error":"image_unavailable"}` - Not a coffee bag or nothing legible: `{"schemaVersion":1,"error":"unreadable_photo"}` ## Rules - Extract only what is legible; never invent or look up. Missing or illegible: `null`. A partial read is not an error. - Only inference: missing country and a region found in one country only (e.g. Cerrado Mineiro → Brazil) fills `origin`, with an `issues` reason inferred. - Do not confuse producer and roaster, country and region, roast date and best-before date, tasting notes and ingredients. - `coffee.process`: washed, natural, honey, anaerobic, other. Use other only for a legible process outside the list, with the original name in `notes`. - `coffee.roastLevel`: light, medium-light, medium, medium-dark, dark. - `coffee.flavorTags`: up to 5 flavors stated on the bag. - Language: write `flavorTags` and `notes` in English, translating if the bag is in another language. Name, roaster, variety, region, producer and farm stay exactly as on the bag. - Net weight always in grams: if printed in another unit, convert it and record the printed value in `issues`. Dates as YYYY-MM-DD. - Useful information without its own field goes in `coffee.notes`, at most 300 characters. - Doubts go in `issues`: `field` (e.g. `bag.roastDate`), `reason` (missing, unreadable, ambiguous, unsupported_value, inferred), `rawValue` and `message`. ## Format ```json {"schemaVersion":1,"coffee":{"name":null,"roaster":null,"origin":null,"region":null,"species":null,"variety":null,"process":null,"roastLevel":null,"flavorTags":[],"notes":null},"bag":{"weightG":null,"roastDate":null},"issues":[]} ```

3. Bring the reply back

In your AI, tap "Copy code" on the reply and come back here.

Paste manually

Paste the AI reply here

A reply inside a code block works too. Review data →