← All models · Donut

use case · practical

Practical: extract document fields

Field extraction without templates. Instead of a parser per document type, you keep a list of questions and let Donut answer each one — total, date, number, vendor — building a structured record from any receipt or invoice, entirely on your device.

Drop a receipt/invoice, or click to choose
An invoice from Acme Supplies Ltd A shop receipt A membership form
Selected document

Fields to extract (edit the questions if you like):

Total–
Date–
Number–
Vendor–

Each row is one document-VQA call. Because Donut answers by understanding the page rather than matching a template, the same four questions work across very different layouts — an invoice, a till receipt, a form.

← Basics · Back to Donut · Next: Wild →