Processing a financial agreement made accurate fast easy

How it works

Read every page. Report only what is there.

Herein turns a document into checked, structured data in two steps, scan then parse, with one rule throughout: when it is not sure, it leaves the field empty rather than guess.

1. 01 · Scan

   Every page becomes words with their exact positions.

   Digital pages are read from the file's own text. Scanned pages are turned upright, straightened and read, so columns line up again even in tables with no ruled lines.

2. →
3. 02 · Parse

   The document is mapped and its tables read.

   Code does everything it can without error. AI models take only what code cannot, and every answer must quote words that are found on the page, or it is rejected.

4. =
5. Result · Output

   Checked data, with its evidence.

   Every value carries the page and words it came from. Warnings flag anything a reviewer should see, and a ledger accounts for every figure, so nothing is missed silently.

{
  "document": { "page_count": 312 },
  "facilities": [
    {
      "id": "facility-1",
      "total_commitment": { "amount": "250000000", "currency": "EUR" },
      "commitments": [
        { "lender_name": "Bank A",
          "amount": { "amount": "60000000", "currency": "EUR" } },
        …
      ]
    }
  ],
  "extraction": {
    "schema_version": "1.1",
    "evidence": [
      { "path": "facilities[0]", "page": 214,
        "snippet": "Bank A | 60,000,000" }
    ],
    "warnings": [],
    "ledger": {
      "gaps": [ { "page": 6, "kind": "percentage", "text": "1%" } ],
      "complete": false
    }
  }
}

Already running AI pipelines or agents? Herein slots straight in — call it as a step in your pipeline or as a tool your agent uses, and hand clean, structured JSON on to your models.

Sign in and view our Developer Docs for the full schema.