Privacy
Voxloom transcribes speech on your Mac. Here is exactly what touches the network and what stays local.
Speech-to-text runs locally
Audio is transcribed on your Mac with whisper.cpp. Voxloom has no cloud speech-to-text path.
The one download: models
Whisper model files come from Hugging Face (huggingface.co/ggerganov/whisper.cpp), and only after you choose to download them. There are no silent downloads.
What's saved, and where
- History: transcript text, status, duration, model, timestamp and delivery outcome, saved in
history.jsonin the app's data folder on your Mac. - Transcript retention setting: Off (no transcripts kept), 30 days, or until deleted.
- Word dictionary:
dictionary.json. Settings:settings.json. Both are stored locally. - Recorded audio isn't saved for later. Saved audio retention is not implemented.
Copy fallback
If inserting text isn't safe and clipboard fallback is enabled, the text is copied to your clipboard. Voxloom never types into password fields and never presses Send.
// Open source: verify all of this in the code on GitHub. Questions: open a GitHub issue. Last updated: October 29, 2026.