Reading receipt photos in the browser, with no upload

ai-bookkeeper's receipts card now takes photos: OCR runs in the page with WebAssembly, and the text is pairable with bank lines.

Tools and projects · 1 min read · Source
ai-bookkeeperreceiptsocr

Dext, Hubdoc and Expensify read receipt photos on their servers, for a subscription. ai-bookkeeper now has a Receipt photos control on its receipts card that does it on your machine. tesseract.js runs as WebAssembly in the page, so the image is never uploaded.

Each photo becomes a text block in the receipts box. The digit look-alike repair the CLI uses (l0.62 to 10.62) runs first, and you can fix a misread amount before pressing Read receipts, which pairs each receipt with the bank line that paid it and lists spending with no receipt.

The check: a script draws a receipt on a canvas, feeds the PNG through the control in headless Chromium, and confirms merchant, tax (0.87) and total (10.62) come out and parse. It passes. Limits: that is one clean rendered image, not a creased thermal receipt in bad light, so accuracy on real phone photos is unmeasured. The reader (about 20 MB) downloads from jsDelivr on first use, only then.

More in Tools and projects

All writing · All in this topic · All tools and research