Pull totals, VAT, dates and invoice numbers from PDF invoices — where every value is backed by an exact quote from the document, or honestly left null. No invented amounts. No wrong due date that bites you at an audit. Each result ships with a signed receipt.
Standard AI extraction fills every field — so a hallucinated total or a guessed tax number looks just like a correct one, and someone has to re-check every invoice by hand. proofetch makes the difference visible: proven values carry their source quote; anything it can't find stays null instead of guessed.
Drop in invoice text, pick the fields you need, and see each value come back with its source quote — or an honest null.
Free demo · text only · no signup. Unprovable fields come back as null, never guessed.
Hit “Run extraction” — each field comes back proven (with its source quote) or honestly null.
You send a JSON schema with the fields you want — proofetch returns each one proven or null. Common invoice fields:
curl -X POST https://proofetch-vektoris-ai-workflows.vercel.app/extract \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"source": { "type": "pdf_url", "value": "https://example.com/invoice.pdf" },
"targetSchema": {
"type": "object",
"properties": {
"invoice_number": { "type": "string" },
"total_gross": { "type": "string" },
"vat_amount": { "type": "string" },
"due_date": { "type": "string" }
},
"additionalProperties": false
}
}'{
"data": {
"invoice_number": "R-2024-0815",
"total_gross": "1,190.00 EUR",
"vat_amount": "190.00 EUR",
"due_date": "2024-09-30"
},
"fields": [
{ "path": "invoice_number", "verdict": "PASS",
"provenance": { "quote": "R-2024-0815" } },
{ "path": "vat_amount", "verdict": "PASS",
"provenance": { "quote": "VAT (19%): 190.00 EUR" } }
],
"verdict": "PASS",
"receipt": { "v": 1, "jobId": "pf_…", "outputHash": "sha256:…" }
}€0.02 per verified field — a typical 20-field invoice is about €0.40, and fields that can't be proven cost nothing. Start with 50 free fields (no card) and run it on your own invoices first.
Normal extraction fills every field — even when the value isn't on the invoice. A hallucinated total looks exactly like a real one. proofetch returns each field with an exact quote from the invoice, or an honest null. You never get an invented amount, date or tax number.
You get null plus a FAIL verdict for that field — never a guessed value. Unprovable fields also cost nothing, so you only pay for data you can actually trust.
Any fields you define in your JSON schema: invoice number, net and gross totals, VAT amount and rate, invoice and due dates, tax/VAT ID, vendor and buyer, IBAN, PO number, currency, and line items.
PDF invoices by URL or base64, plus web URLs, raw HTML, and plain text. Most accounting workflows send a pdf_url.
€0.02 per verified field. A typical invoice with ~20 fields costs about €0.40, and only fields that pass the proof are billed. There's a free tier (50 fields, no card) to test it on your own invoices.
Yes. Every result is sealed in a signed receipt (Ed25519) with input/output hashes, and each value is traceable to the exact passage it came from — useful for audits and disputes.