On-Device AI Auto-Categorization
A 1.7B-parameter model runs on your phone and suggests categories and tags for new transactions — your statements never touch a server.
Why on-device AI is the only AI that protects your statements
Cloud “AI” budgeting apps stream every merchant string to a remote LLM, which often means OpenAI sees your supermarket habits. Budgie loads the model once, then keeps every inference local — same accuracy, zero data exfiltration.
The system blends two signals: a vector-search lookup against your past categorizations and a small generative pass that proposes a tag. As you confirm or correct, the index updates instantly — fully on-device.
What you get
Qwen3 1.7B Q4 model runs entirely on your phone after a one-time download
Nomic embedding model + sqlite-vec for SIMD-accelerated similarity search
Two complementary signals: vector lookup over your history plus a generative tag suggestion
Every confirmation updates the embedding index instantly — accuracy improves as you use it
Statements never leave the device — no OpenAI, no remote inference, ever
How it works
On first run, Budgie downloads a Qwen3 1.7B Q4 model and a Nomic embedding model directly from the Hugging Face hub. Both are stored in your app sandbox. Inference uses ONNX Runtime + sqlite-vec for SIMD-accelerated vector search.
Three privacy-preserving signals
Embedding similarity — your past categorizations index every new transaction
Amount-pattern recurrence — €4.20 every Tuesday morning is probably your coffee
Merchant-name fuzzy match — handles typos, abbreviations, and translated variants
Frequently Asked Questions
Does the AI work offline?
How big is the model download?
Can I correct the AI's suggestions?
Does Budgie use OpenAI or any cloud LLM?
Related Features
Read More on the Blog
Ready to take Budgie for a spin?
Join the waitlist — be first to try the offline-first expense tracker.