Every receipt app has to solve the same first problem: turn a photo of a crumpled receipt into structured data — a merchant, a total, a date. The difference between apps is where that work happens. For most of them, the answer is "on our servers." For Enceipt, the answer is "on your phone, and nowhere else."
This is the whole reason Enceipt exists, so it is worth explaining exactly how it works.
The privacy problem with receipt apps
A receipt is one of the most revealing documents you own. It says where you were, when, and how much you spent. String a year of them together and you have a detailed map of someone's life and business.
Most expense apps ask you to photograph that receipt and upload it. The image lands on a server, an OCR service reads it, and the extracted data is stored in an account tied to your email. Even when the company is well-intentioned, your financial history now lives somewhere you don't control — subject to their breaches, their policies, and their subpoenas.
Enceipt starts from the opposite assumption: the safest place for your data is the one device already in your hand.
How on-device extraction works
When you scan a receipt, Enceipt runs a three-stage pipeline, all locally:
- OCR. On-device text recognition (ML Kit) reads the raw text off the image. The image never leaves the device — not to us, not to anyone.
- Heuristics. A purpose-built parser scans that text for the patterns that matter: the largest plausible total, a transaction date in any common European or US format, the merchant name near the top. This handles the vast majority of everyday receipts instantly and offline.
- Optional AI. For messy, faded, or unusual receipts where the heuristics are unsure, you can enable an on-device AI model that reads the text with far more nuance.
The first two stages ship with every install and need no download, no account, and no connection.
The optional Gemma AI model
The third stage uses Google Gemma 3, a compact, state-of-the-art language model that runs directly on your phone. It is entirely optional:
- It is a one-time download of roughly 0.5 GB, fetched once from Launchwick's CDN and stored in app-private storage.
- Once downloaded, it works fully offline — on a plane, in a basement, anywhere.
- Your receipt text is never sent to a server for AI processing. The model reads it on-device, the same as the rest of the pipeline.
- You can delete it any time from Settings to reclaim the space; Enceipt falls back to its built-in text recognition.
Gemma is provided by Google under the Gemma Terms of Use.
Multilingual by design
Receipts don't all look alike. A German receipt writes the total as 12,90 € and the date as 03.06.2026. A French one uses Montant; a Spanish one, Importe; a Turkish one, Toplam. Enceipt's parser understands these conventions — European decimal commas, thousands separators, and DD.MM.YYYY dates — and the app itself ships in English, German, Turkish, French, and Spanish.
That means the right total is extracted whether the receipt says $162.98 or 162,98 €, and the suggested category arrives in your language.
What we promise
The promise is simple, and the architecture is what backs it up: your receipts stay on your device. No account, no cloud sync, no server reading your data. The optional AI model doesn't change that — it just makes the on-device extraction smarter. You can read exactly what we do and don't touch in our Privacy Policy.