Documents and a JSON Schema in. Validated JSON out.
Send PDFs, scans, Office files or photos with the schema you want filled. DocumentAI reads every page and returns JSON that passes your schema, with the page and quote each value came from.
{ "invoice_number": { "type": "string" },
"total": { "type": "number" },
"due_date": { "type": "string",
"format": "date" } }{ "invoice_number": "INV-1043",
"total": 4812.5,
"due_date": "2026-11-01" }From a folder of documents to data you can load.
- 01
Describe the result
Write a JSON Schema for the data you want: fields, types, lists, required values. Add plain-language instructions if a field needs them.
- 02
Send the documents
Upload in the Playground, or POST a job to the API with your key. One job can take several documents.
- 03
It reads every page
Text PDFs are read from their text layer; scans and photos go through text recognition. Then the model fills your schema.
- 04
Get validated JSON
The answer is checked against your schema before you get it, and each value carries the page and the quote it came from.
Answers you can check.
Any domain: invoices, certificates, forms, contracts, reports. The schema decides the shape; you do not train anything.
Validated against your schema
You get JSON that passes the schema you sent, or a clear error. No free-text answers to parse.
A source for every value
Each value points to the document, the page and the quote it was read from, so a reviewer can check it.
Scans, tables and photos
PDF, DOCX, PPTX, XLSX, PNG, JPEG, TIFF, HTML, TXT and Markdown, up to 50 MB each.
Your own model key, if you like
Connect your own model provider and pay us only for reading the pages; your provider bills the model.
Per page, from prepaid credit.
No subscription, no seats, no minimum spend. Every price includes the model, reading the pages, schema checks and a source for every value.
| Per 1,000 pages | Text PDFs | Scans and images | Own key: text | Own key: scans |
|---|---|---|---|---|
| StandardOur default model. Schema checks and citations included. | $4 | $6 | $1.50 | $2 |
| AccurateLarger models for hard layouts and long tables. | $15* | $20* | $1.50 | $2 |
* Accurate tier: or 3x our measured cost, when that is higher. With your own key, your provider bills the model.
Starter credit. $15 for every new account: about 3,750 text pages, up to 500 pages a day on free credit.
Failed jobs are free. Each job is charged in whole cents, at least $0.01. A dense page (more than 16,000 characters) counts as 2. Full price list
Scores in public, with their price.
We run DocumentAI on public document datasets and publish the score of every model next to its price per 1,000 pages, so you can pick the trade-off.
Open the benchmark →- CORD v2
- FUNSD
- SROIE
- Kleister NDA
- Kleister Charity
- VRDU ad-buy
- VRDU registration
The details.
What do I send, and what comes back?
Documents plus a JSON Schema (and optional instructions). You get back JSON that validates against that schema, with the page and quote each value came from. Jobs run in the background: poll the job or give a webhook URL.
What counts as a page?
One PDF page or one image. A page with more than 16,000 characters of text counts as 2, and very long answers add a page per 2,000 tokens. Text pages are read from the PDF's text layer; scans, photos and Office files need text recognition and are priced as scans.
Which models does it use?
The Standard tier runs our default model; the Accurate tier runs larger models for hard layouts and long tables. With your own key you choose the model and your provider bills it.
How is it billed?
Pay as you go from prepaid credit, per page, at the prices above. No subscription and no minimum spend. A job that fails is not charged; each job is charged in whole cents, at least $0.01. Top up from $10. Every new account starts with $15 of credit, shared with the other Enkiube products.
Is there an API?
Yes. Everything the Playground does is an HTTP API with API keys (available after your first top-up; the Playground works on free credit). The API documentation is in the app.
How accurate is it?
We publish our scores on public datasets (CORD v2, FUNSD, SROIE, Kleister NDA, Kleister Charity, VRDU ad-buy, VRDU registration) with the price per 1,000 pages of each model, on the benchmark page. Your own documents are the real test: try them on the free credit.
Try it on your own documents.
Sign in at documentai.enkiube.com with your Enkiube account. New accounts start with $15 of credit.