What local-first voice means in Side Whisper
A practical look at model downloads, on-device transcription, the local Dictionary and Snippets pipeline, and the limits of the privacy claim.
The model arrives; the recording does not leave
Side Whisper downloads a user-selected Whisper model, records microphone audio for the active dictation session, converts it locally, and runs transcription through the bundled native pipeline. The recording and transcript are not sent to the Side Whisper account backend.
An internet connection is still needed to download a model. Calling the feature local-first does not mean the whole product is offline; it describes where dictation data is processed after that model is available.
Personal text stays part of the local workflow
- Dictionary entries correct names, acronyms, and recurring recognition mistakes after transcription.
- Snippets expand a short trigger into a longer saved phrase before the final paste step.
- Both collections can be imported and exported as explicit local JSON files.
- The final text can be copied or pasted into the application that already owns keyboard focus.
The boundary matters more than the label
Account identity, consent, device, entitlement, and release metadata use the online account service. Browser sites naturally receive data you submit to those sites. Local voice processing does not turn unrelated web activity into offline activity, and the public privacy policy documents those separate boundaries.