No cloud reading
The receipt is read in memory on the device. Reading it needs no server, not even Money Koi’s own.
The receipt scanner
Money Koi reads paper receipts and turns them into clean, categorized bookings — right on the phone. No cloud service analyzes them.
Read on the phone · No cloud service
Why it's different
Most scanning apps upload the receipt photo to a server and have it read there. Money Koi doesn’t. Text recognition, parsing and categorization all run on the device — the exact same scan works in airplane mode.
Typical scanning app
The receipt photo is uploaded to a server, processed in the cloud, and the result is sent back.
Money Koi
Capture, recognition, categorization and saving happen on the device. In a Cloud book, only the finished booking is synced — plus the photo, if photo upload is on.
The receipt is read in memory on the device. Reading it needs no server, not even Money Koi’s own.
The entire pipeline runs with zero connectivity — perfect for a subway or a flight.
With a connection, the app may look up a shop’s VAT number in the EU’s VIES register to name the shop, or open the link that some receipts carry in their QR code. Neither sends the photo or the items.
Under the hood
Every scan runs through a clear, predictable pipeline written by hand — precise, explainable and fast.
Snap the receipt, or pick photos from the gallery. A long receipt is taken in parts.
On-device text recognition (Google ML Kit) turns the image into text, with fuzzy matching to correct common recognition slips.
Merchant, total and date lock in once two readings agree. Austrian receipts with a QR code hand them over at once.
Own parsers for SPAR, BILLA, LIDL, HOFER, ADEG, TEDi, Smyths and OBI — plus a general one for every other shop — pull out each item and its price. The parts of a long receipt are joined at the seams.
Discounts, deposits and returned empties are folded into the right lines. Then the items are checked against the printed total.
Each item gets a category from past corrections, a keyword table, the booking history and fuzzy name matching — all on the device. An item without a match is marked, never filed under “Other”.
Everything shows before saving, and every field can be fixed — the total too. Lines merge into one booking or one per category. Nothing is saved without confirmation.
Trust the numbers
Every scan is checked against the printed total. Saving waits until the items add up and every line has a category — nothing disappears into “Other”.
Each round of improvements is measured on more than 150 real receipts.
What it reads
The scanner is in beta. Merchant, total, date and photo work on most receipts; line items are tuned for eight shops.
Each item becomes its own booking.
Items as a best guess from the general parser, checked in the review.
One booking for eating out — with the photo, or without.
DIY stores, furniture, pharmacies, drugstores and petrol stations: one booking in their category.
In a book kept in yen, won or another currency without cents, the scanner says so before the camera opens.
No setup
The scanner works the moment the app is installed — fully offline, with nothing to set up. Categorization gets sharper as it learns from corrections.
The stack
Fast, accurate text recognition that runs on the device.
Look-alike characters (0 and O, 1 and I, 5 and S) and cut-off names such as “KAROTTENS.” are repaired before parsing.
Store-specific parsers, keyword tables and past corrections map each item to the right category.
The VAT summary under the total is read too. A row only counts once its figures add up.
Totals use precise decimal arithmetic — no floating-point drift.
Fixed names and categories are remembered on the device — quick add uses them too.

The scanner is part of the free budget book. The first receipt is scanned in seconds.