Extraction endpoint

Updated 2026-08-23

Endpoint#

POST https://api.docxtract.io/v3.1/documents
Authorization: Bearer {api_key}
Content-Type: multipart/form-data

Note

There is no /api prefix. api/ is the server's document root, not part of the public path. https://api.docxtract.io/api/v3.1/... returns 404.

Request#

Field	Type	Required	Description
`file`	file	Yes	PDF, JPG, or PNG. Max 10 MB, max 150 pages.
`options`	string	No	JSON-encoded options — see Options

options is a JSON string in a form field, not nested multipart data:

-F 'options={"model":"invoice","store_db":true}'

Response#

{
    "success": true,
    "data": { "vendor": "ABC Suppliers Pvt Ltd", "total": 5900.00 },
    "processing_time_ms": 5577,
    "model_used": "invoice",
    "pages": 1
}

With store_db: true the response also carries "extraction_id": "abc123".

Metadata sits at the root level beside data, not nested under a meta key.

Field	Notes
`data`	Extracted fields. Shape varies by document type — see below
`processing_time_ms`	The AI call duration where measurable, otherwise total wall clock
`model_used`	Prompt template actually applied
`pages`	Page count discovered in the result
`extraction_id`	Stored-result id. Present only when `store_db` is true, which is not the default

Note

The field is extraction_id, not id. Older documentation said id; that was wrong.

document_type and confidence are not returned at root in v3.1, though v3 did return them.

The `data` object#

data is an open map whose shape depends on the document type. Only the envelope is stable.

invoice → vendor, GSTIN, invoice number, totals, line_items[]
passport → name, passport number, date of birth
bank_statement → account details and a row list

Write your integration to read the fields you need defensively rather than assuming a fixed schema. A prompt template can be revised, which can add or rename fields.

Two response codes for success#

Status	Meaning
`200`	Extraction complete. `data` holds the result.
`202`	Document was split for multi-page processing. `data` holds a chunk manifest, not a result.

Warning

A PDF over 3 pages returns 202, and you cannot opt out. Code that checks only success === true will treat the chunk manifest as extracted data and appear to work while returning nonsense. Always branch on the status code, or use an official SDK, which handles both paths in one call.

Errors#

{
    "success": false,
    "error": {
        "code": "invalid_file_type",
        "message": "Unsupported file type. Accepted types: PDF, JPG, PNG",
        "details": { "uploaded_type": "image/gif" }
    }
}

Branch on error.code — it is stable. Message wording is not. See Error codes.

Something wrong or missing on this page? Tell us.