Skip to content

FAQ

Frequently asked questions

If your question is not here, the security page covers how we handle data in more detail, and the API reference covers everything programmatic.

How long do you keep my statements?

Uploaded PDFs and the data extracted from them are deleted one hour after upload, automatically. Nothing is written to a permanent database, and we never keep a copy for training or analytics.

Do I need an account?

No. Anonymous users get 50 free pages a day with no sign-up — metered in pages rather than files, so a one-page statement costs one. An account only becomes useful when you want higher limits, batch uploads or API keys.

Is my financial data used to train models?

Never. Your documents are processed to produce your output and then deleted. They are not used for training, not sold, and not shared with third parties.

What if my statement is password protected?

Enter the password in the prompt that appears next to that file. The document is unlocked in place — the rest of the batch is untouched, and anything you have already corrected stays corrected. Each file gets its own prompt, so a batch with different passwords is fine. The password is used once, in memory, and is never stored or logged.

How accurate is the extraction?

On clean text-based statements from supported banks, typically above 99% at the row level. Every row carries a confidence score, rows below 90% are flagged for review, and the balance check tells you whether the whole statement reconciles against its opening and closing balances.

What does the balance check actually do?

It adds your opening balance to the sum of every extracted transaction and compares the result to the closing balance printed on the statement. If they match, nothing was missed or double-counted. If they do not, we tell you by how much so you can find the bad row.

Can I fix a row the parser got wrong?

Yes. Every cell in the results table is editable, and you can delete rows entirely. Your edits are what gets exported — the download is generated from the table you see, not the original extraction.

Do you support scanned or photographed statements?

Text-based PDFs, which is what banks issue, work best and are what our accuracy figures describe. Scans without a text layer are not currently supported — we will tell you rather than return quietly wrong numbers.

Which banks do you support?

Over 1,000+ institutions across the US, UK, Canada, Australia and the EU, including every major retail bank. The parser also handles unrecognised layouts generically, so an unlisted bank usually still works.

Can I use this from my own code?

Yes — the same conversion runs behind a documented REST API. Post a PDF, poll the job, download the format you want. See the API reference for the full contract.

Still stuck?

Send us the bank and a description of what came out wrong — not the statement itself.

Contact support