Run it yourself

Sundry is MIT licensed, and running it on your own machine is free. It stays free — self-hosting is the free tier.

Two containers and one folder of data:

docker compose up --build

Then open http://localhost:8847. There is no config file to write and no database to create. To reach it from the other devices in the house, use the machine's LAN address on the same port and set a password first.

Your data

Everything sits in ./data beside the compose file: one SQLite database and your receipt photos. Stop the app, copy the folder, and you have a backup. The app also exports to CSV and .xlsx when you want a file to hand someone.

The variables that matter

• APP_PASSWORD — the password gate. Unset, the API is open, which suits localhost and nothing else. Set it on anything the network can reach.
• AUTH_SECRET — an independent signing key for the login token. Without one, tokens are signed with the password itself.
• DB_PATH — where the SQLite file lives. Defaults to data/expenses.db.
• RECEIPTS_ENABLED — receipt upload and OCR. Turn it off on any instance a stranger can open without a password: the text is read on your own CPU, and an open endpoint is a free compute service for whoever finds it.
• TESSERACT_LANG_PATH — a local copy of the OCR language data. See the next section.

The full list, with defaults, is in backend/.env.example and docs/DEPLOYMENT.md.

No outbound requests

Polish and English language data ships inside the image, so a receipt scan reads the text on your own machine and calls nobody. Set RECEIPT_OCR_LANGS to a language the image does not carry and it fetches the whole set from a public CDN instead — point TESSERACT_LANG_PATH at your own folder of .traineddata files to keep it offline. The photograph never leaves the machine either way.

The source

MIT licensed, on GitHub: ptrio42/sundry.cash.

Or do not run it

A hosted instance is planned: your own address, nothing to run. It is not available yet — hello@sundry.cash if you want to know when it is.