Platform · 02

Schema in plain English. JSON out.

No templates. No training. Define the fields you need — including nested tables — and the engine returns structured JSON with per-field confidence.

POST /v1/extract
{
  "document_url": "https://...",
  "schema": {
    "vendor_name": "string",
    "invoice_number": "string",
    "due_date": "ISO date",
    "line_items": [{
      "description": "string",
      "quantity": "number",
      "unit_price": "number"
    }],
    "total": "number"
  }
}

Tables & line items

Multi-page tables, merged cells, currency normalization — handled.

Layout agnostic

Works across thousands of vendor templates without per-format work.

Per-field confidence

Token-level probabilities surface uncertainty so the Review Inbox can catch it.

Try it on a sample