Pull the tables out of a PDF and paste them straight into a spreadsheet. Invoices, bank statements, reports.
Nothing is uploaded. The file is read inside this tab.
Raise it when one row is being split in two. Lower it when two rows are merging into one.
The blank space, in points, that starts a new column.
Your spreadsheet only adds up numbers written the way it expects. A statement from another country usually is not.
Click the row where the table starts — the columns are counted from there and the lines above are left out. Click a column letter to drop a column — the lines above it (titles, addresses) are left out when you copy.
Copy, then paste into Excel, Numbers or Google Sheets — the columns land in place.
One moment — the PDF engine is loading.
Why in the browser: a bank statement or a client invoice should not be uploaded to a stranger's server to become a spreadsheet. Reads: text-based PDFs. A scan with no text layer needs OCR, which this does not do.
Every other converter asks you to upload the PDF to their server. Tablecut does the reading here, in this tab, with your browser's own PDF engine. Nothing is sent anywhere — you can turn off your internet after the page loads and it still works.
That is the whole point. A bank statement, a client invoice, a payroll sheet: those are exactly the documents you are not allowed to hand to a company you have never heard of — and exactly the ones people end up retyping by hand.
The table on screen is what you are about to paste. Check it before you paste it, fix a row, drop the lines above the header.
A converter that quietly drops a digit is worse than no converter at all, because you find out three weeks later when a total does not match. Tablecut would rather show you a messy table than a wrong clean one.
If your own system produced the data, you do not need this. Export a CSV and go. This exists for documents that arrive from outside.
Twenty public documents, none of them ours: US tax forms, research papers, and bank statements published by the banks themselves — including a Société Générale specimen in French, the Banco de Portugal official example, and real Brazilian statements published by a state treasury. Across 916 extracted rows, 90% come out with the columns separated, and 1.8% of cells still hold two numbers that belong apart. A scanned statement in the set is refused, which is the right answer.
Those are the real numbers, including the bad one. The test set and the scripts are published here, so anyone can repeat the measurement instead of believing this paragraph.
Free, and the whole thing. No account, no sign-up, no limit, and your files are never uploaded.
There is also an offline copy — one file with the PDF engine inside it, so it opens on a machine that has never been on a network. It is not on sale right now: the delivery is not wired up yet, and selling something that does not reach the buyer is worse than selling nothing. When it is ready it will be here.