Guide
How to turn a bank PDF into a spreadsheet without uploading it
The usual path is an “online converter”: you send the statement to someone else’s server, wait for an email, and hope they delete the file. That is a bad fit for a document that lists every card payment of the month. Ledger is the other path. The PDF you already have from the bank is parsed in this tab. You check a table. You download Excel or CSV. You close the tab.
The problem with upload converters
A statement PDF is small enough to email, which is why upload sites love it. It is also a complete picture of spending. If the converter is free, you are paying with the file or with a watermark. If it is paid, you still transmitted the document. Ledger does not take that trade. There is no API that accepts a multipart PDF. Ads do not get the bytes either.
What runs in the browser
Mozilla PDF.js (Apache-2.0) loads a worker from this origin and returns positioned text items. Ledger clusters those items by vertical position into lines, splits lines on horizontal gaps, and scores columns: date-like tokens, amount-like tokens, leftover description. SheetJS (Apache-2.0) writes an .xlsx file in memory. file-saver triggers the download, including on iOS. We do not clone a “300 banks” SaaS. Banks change templates; remapping a header is faster than pretending we ship a parser per institution.
Steps
- In online banking, download the statement as PDF — not “print to paper”, not a screenshot. Confirm you can select a line of text in a PDF reader. If you cannot select text, Ledger v1 will refuse the file.
- Open Ledger on this site. Drop the PDF or use the file picker. You may queue up to three files; start with one month while you learn the preview.
- Wait for page progress. A twenty-page statement is normal. Do not background Safari on an old iPhone until the table appears.
- Read the first rows. If “Particulars” is sitting under Debit, change the column role. Ignore columns that are not transactions (page numbers, branch codes) by setting them to Ignore.
- If amounts look positive when they should be withdrawals, enable “Treat (123.45) as negative”. That matches many UK and Commonwealth layouts.
- Download Excel for a workbook with a Notes sheet, or CSV if you are pasting into Google Sheets. The CSV includes a UTF-8 BOM so Windows Excel does not scramble names of payees.
- Compare a handful of rows and the last balance to the PDF. Then send the sheet if that was the job. Delete the download when you no longer need it.
If it fails
- Scan or photo — no text layer. Re-download a digital statement.
- Password prompt in the bank PDF — use the Locked PDF page and type the password in this tab. We do not emit an unlocked.pdf.
- Only a logo extracts — the file is likely image-only even if it is named .pdf.
- Columns shift on page 4 — remap, or split the PDF in the bank portal and drop the month again.
- Tab crashes — the file is too large for the phone. Use a desktop browser or a shorter date range.
After you close the tab
There is no server copy to request or delete. We never received the file. That is the product, not a slogan. The spreadsheet on your disk is now an ordinary file you control.
Compared with the bank’s own CSV export
Many banks already offer CSV. Use that when it exists and when you trust the columns. Ledger is for the case they only gave you a PDF, or the PDF is what your accountant asked for and you still want a table. It is not a replacement for the official document. It is not Photos.app, not Excel’s PDF import (which also stays local, but fights layouts), and not a website that watermarks row 1.
What “text layer” means in practice
Open the PDF in any reader and try to highlight a payee name. If the letters select, PDF.js can see those glyphs and Ledger can cluster them. If the page behaves like a photograph — you can lasso a rectangle of pixels but not a word — there is nothing to parse. Some banks email a “PDF” that is only a scanned printout. v1 will say so instead of filling Description with random fragments from the logo.
Two-column layouts and footers with “Page 3 of 12 / continued” are why we score rows instead of dumping every line. A running balance on the right is kept when it looks like an amount column. A marketing banner in the header is ignored when it does not sit on the same baseline as dates. You still look at the preview. Software that skips that look is how people file the wrong year.