Supported documents — not universal PDF understanding

Supported within tested layouts

Text-based contracts, purchase orders and invoices with recognised document identifiers, explicit fields and supported table headers. The alternative table path requires named columns separated by multiple spaces or pipes. UK day-first, ISO and full English month dates are supported in that path. GBP/USD/EUR amount symbols are recognised where documented; units are preserved literally, never assumed equivalent. Decimal values are not rounded by extraction.

Review or reject

Missing values, conflicting identities and unfamiliar structures may require correction or rejection. You must inspect and confirm all three extracted documents before any audit runs. Review cannot repair missing table rows or a completely unsupported layout.

Not currently supported

Scans/image-only PDFs, arbitrary reading order, heavily merged/wrapped tables, or commercial clauses present only in prose outside the model. No AI or OCR is used. A clean report does not prove all contractual obligations were checked. Findings are review observations, not confirmed recovery.

Uploads are limited to 20 MB per request by default. PDFs are limited to 100 pages and 1,000,000 extracted text characters. These limits do not guarantee every PDF will process.