Day to day
Categories that fill themselves in
Every transaction arrives classified. You only correct the misses.
How the app categorizes
Categorization uses a 5-layer pipeline, in priority order:
- File metadata — if your bank includes a category in the CSV, it's used directly
- Your rules — ones you created manually or approved
- Merchant database — 440+ known merchants by country
- Learned patterns — corrections you made before (≥2 times = pattern)
- Keyword classifier — last-resort heuristic
Default categories
The app comes with 36 hierarchical categories (14 parents + 22 children). You can use them as-is, customize them, or start from scratch.
Automatic rules
When the app detects you always categorize the same merchant the same way (≥3 times, ≥80% consistency), it suggests creating a rule. Rules are created disabled — you approve them before they apply.
Machine Learning (on-device)
Three models trained with 3,197+ samples and 7,680+ phrases:
- Merchant classifier → identifies parent category and subcategory
- Phrase classifier → understands full descriptions in natural language
- Entity extractor → detects amount, date, merchant, and account from free text
Everything runs on your device. No data ever leaves it.
Financial DNA
The app observes each category's behavior and builds a probabilistic profile:
- Is it a fixed or variable expense?
- Is it essential or discretionary?
- How much do you typically spend?
- Is the pattern changing?
If it detects a category isn't behaving as you classified it, it alerts you with an insight.