Receipt → Expense Report

Photograph a stack of receipts, review the guessed vendor, date, and total for each one, and export a clean expense CSV — with nothing ever uploaded.

Drop in photos of receipts and get back one reviewed, correctable CSV — vendor, date, amount, and category per receipt. This tool runs a real OCR engine and regex heuristics entirely offline in your browser to guess the fields, but every guess is editable and flagged clearly when it's likely wrong; nothing exports until you've checked it.

What it does

Receipt → Expense Report reads a photo of a paper receipt, guesses the vendor, date, and total using an in-browser OCR engine and pattern-matching heuristics, and puts every guess in front of you as an editable field before anything is exported. Drop in one receipt or twenty at once; each is scanned in place with real progress, low-confidence guesses (an ambiguous date, a receipt with no clear "total" label) are visually flagged with a plain-language explanation of why, and a raw-OCR-text view lets you double-check any guess against what the engine actually read. Check each receipt off as "Reviewed" once its fields look right, and export every reviewed receipt as one clean CSV — date, vendor, amount, category, and which photo it came from.

Who it's for: a freelancer, contractor, or small-business owner who has a stack of receipt photos and wants a clean expense CSV out of them without retyping every line by hand — especially someone who currently hands a bookkeeper a shoebox (or a phone full of photos) and pays for the data-entry time.

Example scenario: a contractor wrapping a two-week remodel job has photographed 18 job-site receipts on their phone. They drop all 18 into this tool, spend about 15 minutes reviewing the guessed vendor/date/amount for each one (fixing the two or three the tool flagged as low-confidence), assign a category to each, and export one CSV to send their bookkeeper — instead of a bag of curled paper receipts and 40 minutes of someone else's billable data-entry time.

What this does not do: it is not fully automatic bookkeeping, and it does not claim to be. Vendor, date, and total are OCR-assisted guesses from regex heuristics — a receipt with a subtotal/tax/tip/total breakdown and no clear "total" label is a genuinely hard case, and this tool's own testing found a real example where its best guess was wrong (documented honestly in the product's own build notes). Every receipt requires a human review step before export; nothing is exported without it. It does not read handwriting reliably, does not pre-process images (no auto-crop, no deskew, no contrast fix), only reads English text, and does not replace a bookkeeper or accountant's judgment — it replaces the typing.

Features

• Scans one or many receipt photos in a single batch, each with its own live OCR progress bar.
• Guesses vendor, date, and total per receipt, with low-confidence guesses visually flagged and explained in plain language (not just silently presented as fact).
• Every field is editable before export — nothing is exported until you check "Reviewed" on that receipt.
• A raw-OCR-text view per receipt lets you double-check any guess against what the engine actually read.
• Runs entirely offline after the file loads — zero network requests during scanning, verified directly across Chrome, Safari, and Firefox, not just claimed.
• Exports one CSV (date, vendor, amount, category, source image) built from your reviewed, corrected values.

Browser support

FAQ

Does my receipt photo get uploaded anywhere? No. Every image is decoded and scanned entirely inside your browser tab. This was verified directly with network-request interception during testing across three browser engines — zero requests leave the page during OCR, on any image.

How accurate is the extraction, really? On clean, well-lit receipts with an unambiguous "Total" line, real testing found vendor, date, and amount all correct. On a receipt with no clear "total"/"balance"/"amount due" label — for example one where the only visible numbers are line items plus "Cash Tendered" and "Change" — real testing found the tool's fallback guess (the largest dollar amount on the receipt) was genuinely wrong, though it was correctly flagged low-confidence rather than presented as fact. Rotated and blurry photos (simulating a real phone photo) still extracted correctly in testing, but engine confidence measurably dropped. Always read the flagged fields, and ideally every field, before exporting.

Do I have to review every receipt, or can I trust it blindly? You have to review each one — this is by design, not a limitation to work around. Nothing is included in the exported CSV until you check "Reviewed" on that specific receipt; unreviewed receipts are silently excluded from export (and the tool tells you how many were skipped) rather than exported with unverified guesses.

Does this replace my bookkeeper or accountant? No. It replaces the typing, not the judgment — categorization decisions, tax treatment, and the actual bookkeeping are still yours or your bookkeeper's to get right. This tool's job ends at "here's a clean CSV," not "here's your books."

Can I use this for client or commercial work? Yes. One-time purchase, personal and commercial use included.

Is there a subscription? No. One-time purchase, yours to keep and reuse, with no ongoing cost — unlike every cloud-based competitor found during this tool's own research, which all charge monthly or annually.

Does this work on my iPhone/iPad? Not verified. It's built with standard web APIs and a small-viewport layout was checked via emulation, but real iOS/Android testing wasn't possible on this machine (see Platform support above) — treat mobile as untested, not promised.

What happens if I lose the downloaded file? Nothing is stored anywhere but your own device — there's no account or cloud copy to recover it from, so keep the file itself somewhere safe (the same as any other file you buy once and own).