Using the workbench.

1. Add a small batch

Enter a client label, then choose or drop PDF, JPG or PNG files. You can change each document’s client later. Try sample files to practise with a fictional invoice and receipt; samples use the same reading and review process as your files.

• At most 25 files per batch.
• At most 10 MiB per file and 50 MiB across the batch.
• PDFs may contain up to 5 pages. Password-protected PDFs are unsupported.
• Use English printed text. Handwriting, HEIC and multiple separate receipts combined into one PDF are not supported.

Scans and images require OCR. Reading resources must load before they can be processed. Large or unclear images may need to be resized or rescanned. A failed file stays in the queue; it does not disappear from the batch.

2. Review every document

Select a document and compare the suggested merchant, document date, total and currency with the original. Check all pages. Use the client label you want on the output folder.

Use the invoice date rather than the payment due date, and the document total rather than the unpaid balance. Dates such as 03/04/2026 need a human decision. A dollar sign alone does not identify USD, CAD or another dollar currency. A missing or uncertain field remains for you to fill in.

Click Confirm reviewed details after checking. Editing any field removes its confirmation. Confirmation is your review decision, not a claim that OCR was accurate.

3. Download the handover

Every document must be confirmed or explicitly excluded before downloading. The ZIP contains:

• Byte-for-byte copies, arranged by client and document month.
• A CSV list linking the original name to the new path, reviewed fields and SHA-256 checksum.
• A short README explaining the export.

Example path: Client / 2026-10 / 2026-10-18_Merchant_18.75_USD.pdf. Invalid filename characters are cleaned; conflicting names receive a numbered suffix. Distinct files are not silently overwritten. A repeated file stays visible so you can decide whether to exclude it.

The CSV protects text fields from spreadsheet formulas. A leading apostrophe may appear in formula-like text. Validated amounts remain decimal numbers, including negative credit notes. The CSV is a handover list, not a guaranteed import format for accounting software.

If something needs attention

A file could not be read

Read the message, then retry if the issue was temporary. If the file passed format and size checks but OCR failed, timed out or was cancelled, an Enter fields manually button is available. Use it after viewing the original, then enter and confirm every field. Unsupported, password-protected, oversized or damaged documents cannot bypass the checks: make a new supported file outside this tool, or explicitly exclude the old file before exporting the rest.

The fields look wrong

Correct them from the source. If the source cannot be read reliably, exclude it and resolve it separately. Do not assume an OCR suggestion is correct.

Reading seems slow

The first OCR use downloads reading resources. Smaller batches help on slower devices. Cancel stops remaining work; cancelled files can be retried.

I need to start over

Clear this batch releases the document previews and working records. Refreshing or closing the tab also clears your edits. Download first if you need your reviewed work.

Scope and responsibility

Your original files are not renamed or edited. You receive a new ZIP, which is not encrypted. There are no accounts, payments, automatic bookkeeping entries or tax calculations in this beta. Always review the source and use the appropriate professional process for your accounting records.

Read the privacy explanation for the distinction between downloading application resources and processing your documents.