The Problem
Every budgeting app wants access to your bank accounts, charges monthly subscriptions, and stores your financial data on their servers. I wanted to track expenses without giving away my financial privacy. I also tend to keep a drawer full of receipts that need processing, but they usually end up shredded without ever being recorded. Granular visibility into spending patterns would be useful too.
The Solution
Budget Tracker is a desktop application that runs entirely on your machine:
- 100% Offline: No internet required, no data leaves your device
- AI Receipt Scanning: Local LLM extracts items and prices from receipt photos
- No Subscriptions: One-time download, runs forever
- Cross-Platform: macOS, Windows, and Linux
Features
Expense Management
- Categories and subcategories for detailed tracking
- Split expenses for shared costs
- Store and item price history
- Multiple budget contexts (personal, business, etc.)
Receipt Processing
The standout feature: photograph your receipts and let AI do the data entry.
- Scan QR code to transfer via local WiFi
- Take a photo with your phone
- Local LLM extracts line items automatically
- Review and categorize in seconds
AI Models
| Hardware | Recommended Model | Speed |
|---|---|---|
| GPU 8GB+ VRAM | llama3.2-vision:11b | 10-20s/receipt |
| GPU 4-8GB VRAM | minicpm-v | 30s/receipt |
| CPU only 16GB+ RAM | llama3.2-vision:11b | 60s+/receipt |
Budget Planning
- Monthly allocations per category
- Visual progress indicators
- Spending trends and dashboards
Data Control
- SQLite database (single file, easy to backup)
- Export/import functionality
- No vendor lock-in
Tech Stack
| Component | Technology |
|---|---|
| Desktop Framework | Tauri 2.x (Rust) |
| Frontend | React 18, TypeScript, Tailwind CSS |
| Database | SQLite with Drizzle ORM |
| Local AI | Ollama or bundled llama.cpp sidecar |
| OCR | llama3.2-vision for receipt parsing |
Why Local AI?
Cloud OCR services are convenient but:
- Require internet connectivity
- Cost money per scan
- Upload your financial data to third parties
By bundling llama.cpp, the app works completely offline and keeps your receipts private. The AI runs on your hardware: no API keys, no usage limits, no data sharing.
Open Source
The app is currently undergoing a rewrite from an early MVP originally deployed on Vercel and Supabase. It will be available as a public GitHub repository once complete.
Related Articles
- Budgeting Without Big Brother: Why your financial data should stay on your device