There is no account and no database. Your file is never uploaded. The text of it is: to read the receipt, an AI has to be shown what it says, and that AI runs on this site's own server.
PDFs are parsed by pdf.js inside the tab. A page with no text layer is a scan, so it is rendered to a canvas and read by tesseract.js, also inside the tab. Neither library sends your file anywhere. The page pictures you see, the highlight boxes drawn over them and the PDF you download at the end are all built in the tab too.
The text that pdf.js or OCR pulled out of your pages, and nothing else. Not the file, not the page images, and at most the first 12,000 characters. It goes to this site's own Cloudflare Worker, one short question at a time.
The text of your receipt is read by OpenAI. It goes to our own Worker first, and the Worker asks GPT-5.6 through Cloudflare AI Gateway using our own OpenAI API key. So the words on your receipt do reach a company other than us. Your file and your page images do not: those are read, highlighted and exported entirely in this browser and are never uploaded anywhere.
OpenAI holds that text for thirty minutes, and here is why. Checking one receipt asks nine to fifteen separate questions about it, and every question has to send the receipt again. Rather than pay to send the same receipt fifteen times, we let OpenAI keep it briefly and reuse it across those questions. It is used to answer your questions and nothing else, it is not used to train anything, and it is gone after thirty minutes. That is the one piece of your receipt that outlives its request, and it is the whole of what this page has to warn you about.
Nothing else is written down. We keep no database, no log of your receipt, no analytics on it, and the gateway in front of OpenAI has request logging explicitly turned off, so no prompt is stored there either. The running cost this tool reports comes from the token counts in the reply, not from anything retained. What our server can be asked is fixed too, a short list of questions about a receipt, so the route cannot be used for anything else.
If you would rather no third party saw the text at all, do not use this tool on that receipt. There is no setting on this page that changes it, and we would rather say so than offer you a switch that does not mean anything.
For a while this page could run the AI in your own browser, and you picked which. That is gone. The on-device option meant an 840 MB download before the first check and about a minute per receipt afterwards, to run a model small enough that it once answered "who was paid" with the buyer's own email address. Almost nobody picked it twice, and a choice where one answer is always wrong is not really a choice: it is a question about graphics hardware in front of a question about a receipt. So the page asks for a file and nothing else, and this page says plainly where the text goes.
The exported PDF contains your pages with the highlight boxes drawn on, plus a small record of what was found and which corrections you made, stored where no reader displays it, so that dropping the file back here later restores your work. That record includes the file names and SHA-256 hashes of the pages you combined. It is written to your device by your browser and is not sent anywhere.
This site is not affiliated with, endorsed by, or officially connected to Utah Fits All (UFA), Odyssey, or the Utah State Board of Education. It reads the published receipt requirements and tells you what it can find. It cannot promise an approval.