Turn a receipt into the shop, the date, the lines and the total.
This file is sent to a server so a vision model can read it, then discarded. Nothing is stored and nothing is used to train a model. If that is wrong for a particular receipt, don't use this one.
One or a pile — up to 20 at once. JPG, PNG, WebP, HEIC, AVIF and PDF up to 8MB each; larger images are scaled down on your device before anything is sent.
or
Reading the receipt…
This tool sends your receipt to a server so a model can read it. It is held in memory for the second or two that takes, never written to disk, and never kept afterwards. No receipt sent here is used to train a model. Everything after the reading — the editing, the checks and the export — happens on your device.
Turns a photo, screenshot or PDF of a receipt into fields you can use: the shop, the date, every line that was bought, the tax and the total. Nothing has to be typed twice, and everything on the page can be corrected before you export it. Drop one, or drop the whole pile — up to twenty at a time, each read separately so one unreadable photograph costs you that receipt and not the other nineteen.
Receipt text is the hardest kind to read — thermal print on a creased slip, abbreviations like ORG BAN where a word should be, and numbers in three different columns. So the interesting part is not the reading, it is the checking: the lines are added up against the printed subtotal and the subtotal against the printed total, and anything that disagrees is marked for you to look at. Nothing is corrected automatically. A receipt that does not add up is often a real receipt, and quietly rewriting a figure to make the sums work would hide the one thing worth knowing.
The file is sent to a server to be read, because a model that can read a crumpled receipt does not run in a browser tab. It is held in memory for the second or two that takes and is never stored. Everything after that — the editing, the totals, the export — happens on your device.
Pull the merchant, date, tax and total off a lunch or hotel receipt and paste them straight into the form your employer uses, instead of squinting at a photo in another window.
Download a CSV with one row per purchase and open it in Excel or Google Sheets. The receipt-level totals sit on the first row only, so summing a column gives the right answer.
Thermal receipts fade within a year or two. Reading one into text now means the warranty, the price and the date are still legible when you need them.
Get every line and its price as a list you can hand round, rather than passing one phone photo between five people.
The lines are added up against the printed subtotal and total. If they disagree, the page says by how much — which is sometimes a misreading and is sometimes the shop.
Drop a photo, a screenshot or a PDF — or choose one from your device. JPG, PNG, WebP, HEIC, AVIF and PDF up to 8MB. Photograph the whole slip, including the total at the bottom.
Images are scaled down on your device first, then sent and read. Typically a second or two.
Every field is editable, and anything that does not reconcile is marked with what it was expecting. Fix a number and the checks re-run as you type.
Copy the receipt as text, download a CSV for a spreadsheet, or download JSON if something else is going to read it. Exports are built from what is on screen, so your corrections are in them.
Yes, and there is no account. There is a daily limit on how many receipts one visitor can scan, because reading them costs us money per file — the page tells you how many you have left after each one.
No. It is sent to a server so a model can read it, held in memory for the second or two that takes, and discarded. It is not written to disk, there is no copy to come back to later, and no receipt here is used to train a model. The page says this above the upload box rather than after it, because it is something to know before you drop a file.
Because two numbers the receipt itself printed disagree: the lines do not add up to the subtotal, or the subtotal and tax do not reach the total, or a quantity times a unit price is not the line total. The mark means look, not wrong — the tool cannot tell which of the two figures was misread, and sometimes neither was.
No, and that is deliberate. The printed figures are copied across exactly as they appear, even when they do not reconcile. Silently adjusting one to make the arithmetic work would hide the discrepancy, which on a receipt is usually the only thing worth noticing.
It will try, and how well it does depends entirely on the handwriting. Printed and thermal receipts are what it is good at. Whatever it reads, every field is editable.
Yes — an emailed order confirmation or an invoice saved as a PDF works the same way as a photo. Upload the PDF itself rather than a screenshot of it, since the text is sharper.
Yes — drop up to twenty. Each is read separately and appears in a list beside the fields; click any one to check and correct it. Copy takes the receipt you are looking at, and the CSV and JSON downloads take all of them at once.
One row per purchased line, with the merchant, date, order number and currency repeated on each. The subtotal, tax and total appear on the first row only, so adding up the total column gives the receipt's total rather than a multiple of it.