Pull the fields you name out of any unstructured text as typed JSON, with the evidence for every value.
This data extractor turns unstructured text into structured JSON. Name the fields you want — invoice_number, total, due_date, candidate_name, skills — pick a type for each, and paste the document. You get back a value per field, the verbatim span of the text each value came from, and an explicit not found for anything the document never states.
The type is what makes the output usable downstream. number and integer turn "$1,299.00" into 1299.0; date normalizes "March 5, 2026" or "5 March 2026" to 2026-03-05; list collects repeated items into an array; boolean answers a yes/no question about the text. A value that won't convert to its declared type is reported missing rather than handed back in the wrong shape, and deliberately ambiguous numeric dates such as 03/04/2026 are never guessed.
Every filled field has to quote the text it came from, and that quote is checked against the document — so an invented value is dropped instead of returned. This is extraction, not inference or summarization: if the answer isn't stated in the text, the right result is "not found". For anything that needs reading between the lines, use the summarizer instead. An optional per-field description and the Context box are the way to disambiguate similar fields ("the total due, not the subtotal") without asking the model to invent anything.
Need this in your own product or pipeline? The Extraction API takes the same inputs and returns the same data map, per-field evidence and missing list as JSON — handy for invoice and receipt processing, resume parsing, email and ticket triage, or pulling attributes out of product copy. Pair it with the sentiment and tone utilities, or contact us about document processing at volume.