AWS Textract (AnalyzeExpense) alternative for invoice data extraction
Send an invoice, get the fields you name back as JSON. $0.03 per document, no account, no minimum, nothing stored.
AWS Textract (AnalyzeExpense) and Agent Data Tools side by side
| AWS Textract (AnalyzeExpense) | Agent Data Tools | |
|---|---|---|
| Price per invoice | $0.01 per page for the first million pages, then $0.008 (US West, Oregon) | $0.03 per document with a key; $0.018 on the $9 a month plan (500 documents) |
| Free to start | 100 pages a month for the first three months, for new AWS customers | Free API key by e-mail |
| Commitment | None, but an AWS account, credentials and SDK setup | None: packs never expire, plans cancel any time |
| Long invoices | Per page | One price per file (first 24,000 characters; three pages of a scan) |
| Account before the first call | Yes | No: card checkout gives a key, or an agent pays per call |
| EU VAT numbers checked in the result | Not on its pricing page | Yes, with /invoice/check ($0.05) |
AWS Textract (AnalyzeExpense) figures from its pricing page, read on 14 November 2026. It returns the fields AWS defines for invoices and receipts; you map them to your own names.
When AWS Textract (AnalyzeExpense) is the better choice
- Cheaper per page than we are
- Runs inside AWS next to the rest of an AWS workload
What you get with Agent Data Tools
- $0.03 per document with an API key: one call per file, whatever its page count. It reads the first 24,000 characters of a file, and up to three pages of a scan.
- No minimum and no yearly plan: one-off credit packs that never expire, or a monthly plan you can cancel at any time (from $9 a month).
- No account: pay by card and get a key on the next page, or let an agent pay per call in USDC (x402).
- A free API key by e-mail for a first test.
- Failed calls are never charged, and a file that cannot be read costs nothing.
- You name the fields you want in each request (invoice_number, total, due_date, supplier…): no template, no training. A value the file does not state comes back as null, never guessed.
- PDF, Word, Excel and PowerPoint files, and scans and photos by OCR.
- Files sent in a request are held in memory for that request only: not stored, cached or logged.
- One-call invoice check at $0.05: the fields, plus the seller's and buyer's EU VAT numbers checked in VIES, the IBAN, and whether net + VAT = total.
One request
POST the file and name the fields in the query string:
curl -H "Authorization: Bearer $API_KEY" \
-F "file=@invoice.pdf" \
"https://agent-data-tools.revmesh2074.workers.dev/document/extract?fields=invoice_number,invoice_date,due_date,supplier,supplier_vat,total,currency"Questions
Is Agent Data Tools cheaper than AWS Textract (AnalyzeExpense)?
Not per page at the pay-as-you-go price: AWS Textract (AnalyzeExpense) starts lower per page (inside the $9 a month plan a document costs $0.018). Agent Data Tools is per document with no account, no minimum and no plan, which costs less when volumes are small or uneven, and when invoices run to several pages.
Do I need to train a model or set up templates?
No. Send the file and the field names you want; the model finds them whatever the layout or language.
Are my files stored?
No. A file sent in a request is held in memory for that request only. It is not stored, cached or logged.
How accurate is it?
An AI model reads the file. Check amounts and dates that matter before you act on them; a field the file does not state comes back as null rather than a guess.
Other comparisons: Veryfi · Mindee · Nanonets · Reducto · Parseur · Docparser · LlamaParse (LlamaExtract)