Document to structured data
Drop in a document. Get back rows you can check.
Every field comes back with a confidence score, and anything the model was unsure about is flagged for a person instead of being quietly written through. Download it as CSV.
What happens to your file. It is read into memory, parsed, and dropped — before this page finishes loading. It is never written to disk and its contents are never logged. The extracted values are kept for fifteen minutes so the CSV link works, then deleted. Upload a real invoice if you want; that is what it is for.
What it reads well, and what it does not
- Invoices, receipts, purchase orders, statements, intake forms
- Recognised by their own wording, so the same document always comes back with the same columns. Anything else falls back to a generic field set and says so.
- Digital PDFs — the ones with a text layer
- This is the strong case. Most documents a business retypes are these.
- Scans and photographs
- Read as an image, which is the least reliable input there is. It works, and every field it produces should be checked. Said here rather than discovered later.
- Handwriting
- Do not rely on it. That is not a limitation worth hiding.
Why the confidence column is the product
Any tool can put a number in a cell. This one earns the number: the value has to parse as the thing it claims to be, the date has to be plausible, the line items have to sum to the subtotal, and the subtotal plus tax has to be the total. The model's own confidence is only ever a ceiling — the arithmetic can lower it and never raises it.
A flagged field costs a person ten seconds. A confidently wrong one costs the relationship.
Engines
- Together ready
- GPT-4o mini reads scans ready
- Claude reads scans needs ANTHROPIC_API_KEY
- Local model ready
With no hosted credential this runs on a local model, which reads text layers only. Every report says which engine produced it.
Get this for your business
A build configured to your document formats, your fields and your review threshold: $1,800. Fully local, where nothing leaves your network: $2,600 — the same product, for clients whose documents may not go to a hosted API.