OneScribe reads what you scan and turns it into organized, searchable details, with the AI running on your iPhone or iPad.
OneScribe turns a photo of a receipt, a lease or a boarding pass into a structured card: the vendor, the dates, the amounts, the deadlines. The reading happens on the device: there’s no account to create and no OneScribe server to send anything to. Scanning is free; Pro is a single $9.99 purchase.
Most scans end up as a flat image or a block of text. The information people actually need, like when the warranty ends or what the lease says about notice, stays buried until someone rereads the page. Many of these documents are also the most private ones a person owns: tax forms, prescriptions, IDs. That made on-device processing the starting point, not a feature.
Every document type is a Swift @Generable struct, so Apple’s on-device model fills typed fields instead of writing prose. There are 151 schemas and about 1,420 field instructions, all written in one house style. After each extraction a validator flags impossible dates, negative amounts and balances that don’t add up; a critical issue triggers one retry in a fresh session.
A single Vision document request per page returns the text in reading order with its structure and barcodes, which let the project delete about 1,200 lines of its own layout code. A Core ML classifier identifies the document type first and returns immediately when it’s confident; only ambiguous documents reach the language model.
Document links come from a pure rule engine: same vendor, related types with matching amounts, identical policy numbers. The briefing picks what is due within 30 days, and the model only writes the headline and first paragraph, with fixed text standing by if it fails.
OneScribe is free to download and scan with. Pro is $9.99, once.