Text rules match the labels on the text layer — deterministic and instant, with no model. The AI route reuses the engine behind Ask this PDF: the on-device model, or your own key with the page text going only where you choose. Scanned invoices are read with OCR inside the same run while the daily allowance covers them. You download one ZIP with extracted-fields.xlsx and extracted-fields.csv — one row per invoice, one column per field, the file name first. The whole batch counts as one of your daily jobs, and nothing is uploaded.
How it works
- Stage the invoice PDFs and open Extract fields → CSV
- Name the fields — Invoice number | Invoice No., Date, Total are pre-filled
- Extract and download the spreadsheet — one row per invoice
Questions people ask
What if an invoice is missing a field?
Its cell stays empty rather than being guessed, so the gaps are visible in the table and easy to fill by hand from the file named in the first column.
How does it count against the free plan?
The whole batch is one of your 3 daily jobs, files up to 25 MB, and up to 5 files per run on the free plan — Pro removes the cap. Scanned pages share the 5 free OCR pages a day.
Are my invoices uploaded?
No. Reading, matching and the spreadsheet happen in your browser tab. The AI route with the on-device model also runs on your device; only your own API key sends page text to the provider you chose.