Budgie logo
Budgie

On-Device AI Budget App — Local LLM, No Cloud Inference

Cloud AI assistants for budgeting send every transaction to a remote server for "intelligence". Budgie runs the LLM and embeddings on your phone — your data never leaves.

Why this matters

Cloud AI assistants for budgeting send every transaction to a remote server for "intelligence". Budgie runs the LLM and embeddings on your phone — your data never leaves.

Budgie vs. the category

Qwen3 1.7B for chat-style suggestions and a 768-dim embedding model for nearest-neighbor lookups. Both run via ONNX Runtime on iOS and Android.

Roughly 1 GB combined. The download is one-time, opt-in, and only triggers if you turn on AI features.

A cloud assistant ships your transaction titles to a remote model and trusts the provider's privacy policy. Budgie's models live on your device — there's no provider to trust.

Inference runs in milliseconds for embeddings and a few seconds for the LLM, only when you trigger it. Background impact is negligible.

Feature comparison

FeatureBudgieCloud AI budget assistants
Where AI runsOn your phoneVendor's cloud / remote AI service
What gets sentNothingEvery transaction title, often more
Works offlineYesNo
AI subscription requiredNoOften yes
Privacy from AI providerTotal — no provider existsBound by their privacy policy
Suggestion qualityImproves with your correctionsStatic, plus your data trains their model

Frequently Asked Questions

Which models does Budgie run on-device?
Qwen3 1.7B for chat-style suggestions and a 768-dim embedding model for nearest-neighbor lookups. Both run via ONNX Runtime on iOS and Android.
How big is the download?
Roughly 1 GB combined. The download is one-time, opt-in, and only triggers if you turn on AI features.
How is this different from a cloud AI assistant?
A cloud assistant ships your transaction titles to a remote model and trusts the provider's privacy policy. Budgie's models live on your device — there's no provider to trust.
Does on-device AI drain battery?
Inference runs in milliseconds for embeddings and a few seconds for the LLM, only when you trigger it. Background impact is negligible.

Ready to take Budgie for a spin?

Join the waitlist — be first to try the offline-first expense tracker.