Open the Receipt OCR + Categorizer and drag in one or more receipts — phone photos (JPG, PNG, WebP) or PDFs. When you click Process, the tool reads each receipt locally: a photo is recognised with tesseract.js optical character recognition, and a born-digital PDF reads its embedded text layer with pdfjs. It then parses that text with rule-based heuristics: the merchant is taken from the first substantial line at the top, the date from a date pattern, the grand total from a labelled “Total / Amount Due / Balance” line read bottom-up (with the largest amount as a fallback), and the tax from a “Tax / VAT / GST” line. Each receipt appears as an editable row with a confidence on each field, and the merchant is matched against a keyword map to pick an expense category. Correct anything that looks off, then export.

The Receipt OCR + Categorizer runs completely client-side. Photos are recognised with tesseract.js, digital PDFs are read with pdfjs, and the Excel workbook is built with exceljs — all libraries load from the app’s own bundle, with no server, no API, no AI and no CDN call for your receipts — and the page works offline once cached. The files you upload never leave the page; nothing is sent, logged or stored, and object URLs and the OCR worker are cleaned up when processing finishes or you leave. That is the key difference from most receipt-scanning apps, which upload your receipt photos to their servers (and sometimes use them to train models). Because your receipts stay on the device, the tool is safe for personal and business expenses, and only the Excel or CSV you choose to export ever leaves the page.

Categories follow the buckets receipt-scanning and expense apps use: Meals, Travel, Car & Vehicle (fuel, parking, tolls), Office Expenses, Supplies, Software & Subscriptions, Professional Fees, Utilities, Telephone & Internet, Shipping & Postage and others, plus Uncategorized when nothing matches. The tool matches the detected merchant (and, failing that, the whole receipt text) against a keyword map — for example a coffee shop maps to Meals, a gas station to Car & Vehicle, a SaaS vendor to Software & Subscriptions. The match is a suggestion: every category is a dropdown, so you can override any receipt. If you edit a merchant, the tool re-runs categorization for that row unless you have already picked the category yourself, in which case your choice is kept.

Accuracy depends heavily on the receipt. A clean, born-digital PDF extracts very well because it uses the real text layer, but most receipts are photos of thermal print, which fades (the merchant name at the top fades first), and phone photos are skewed and low-contrast, so OCR misreads characters. The field logic is rule-based, not AI, so a common slip is grabbing the subtotal instead of the grand total when the layout is ambiguous, or missing the tax line. That is why every field is editable, each carries a confidence, and the category is a dropdown. Use it to do the bulk of the data entry and to keep your receipts on your device, but review and correct each receipt — especially the total and tax — before you rely on it for an expense claim, reimbursement or tax filing. It is not AI, not an accountant, and not tax advice.