Voice Transaction Entry
Say “twelve dollars coffee this morning” and Budgie logs it. whisper.rn (whisper.cpp backend) and the on-device LLM both run locally — no audio leaves your phone.
Why every voice budgeting app today is a privacy hole
Voice is the fastest input mode for an expense — but every voice budgeting app today streams microphone data to a vendor server. Budgie keeps the audio stream entirely on the device, then runs whisper.rn (a whisper.cpp-backed React Native library) for transcription and a local LLM for entity extraction.
The result is a pre-filled transaction form: amount, category, account, and merchant — with the same AI category suggestion pill you get for typed entries. Confirm, edit if needed, save.
What you get
whisper.rn (whisper.cpp backend) runs Whisper-small locally for accurate, multilingual transcription
On-device LLM extracts amount, merchant, date, and category from natural speech
Audio never leaves the device — no Siri-style cloud round-trip
Pre-fills the same quick-entry form you would use by typing — confirm or correct
Works during the AI model loading phase too — visual progress indicator built-in
How it works
Tap the mic in the quick-entry sheet. whisper.rn transcribes locally using the whisper.cpp engine. The local LLM extracts amount, merchant, and date hints from the transcription and applies the same on-device category suggestion pipeline used for typed transactions.
Three steps from speech to saved
Tap the mic in the quick-entry sheet
Speak naturally — “twelve dollars coffee at the airport”
Confirm or correct the pre-filled form, then save
Frequently Asked Questions
Which languages does voice entry support?
Is my voice recorded anywhere?
What if Whisper mishears me?
Does it work offline?
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.