Submit a document for extraction
Accepts a single document and queues it for asynchronous extraction.
More details
More details
Authorizations
API key for authentication
Body
The document. Must be application/pdf, image/png, image/jpeg, or image/tiff.
Optional extraction schema as a JSON string. When omitted, the default pipeline generates one and returns it under result.schema with generated: true. The object must match the engine's DocumentSchema shape: required 'document_type', plus optional 'tables' and 'form_fields', where each form field requires 'field_name' (the verbatim label printed on the document), 'field_type' (text | number | date | checkbox | radio | dropdown | signature | handwriting | graph | image | logo | stamp | qr_code | textarea | currency | unknown) and 'page_index' (1-indexed). A schema that parses but does not match that shape fails later on the job, with the engine's validation errors under error.detail.
"{\"document_type\":\"insurance_claim_form\",\"form_fields\":[{\"field_name\":\"Policy Number\",\"field_type\":\"text\",\"page_index\":1}]}"
Optional model-provider hint (e.g. gemini).
Reserved for contracted pipelines. Omit it to use the default pipeline, which generates a schema when none is supplied. An unknown value is rejected with 422.
When true, the result includes per-page OCR output under result.pages. Defaults to false.
Optional HTTPS URL to receive the extraction.job.settled webhook when the job settles. A blank value is treated as omitted.
Opaque caller-supplied reference, echoed on the job and webhook payload.
256