Demo · document extraction

Drop a document. Get JSON.

A scanned page, a photo of a form, a PDF up to five pages. The model reads it and returns parties, dates, amounts, identifiers and every field a person would look up, as one JSON object against a fixed schema. This is Qwen3.8-27B on our own GPU in Helsinki, the same endpoint the paying traffic runs on.

 
Extracted
Raw JSON

      
What happens to the file

It is read once and discarded. Images go to the model straight from memory; a PDF is split into page images in RAM and deleted the moment the request is done. The only thing kept is a counter per address so the limit works: five documents a day, reset at 00:00 UTC. No account, no cookie, no upload log with file names.

The extraction schema is fixed for the demo. Through the API you set your own: the endpoint supports JSON schema, JSON mode and tool calling, with a 262,144-token context and image input. Prices and the trial key are on the docs page.