Privacy is part of the processing path.
Murmr Voice is designed around on-device speech and explicit model choices. This page describes the current product behavior as of August 22, 2026.
Microphone and speech recognition
On Mac, Murmr activates the microphone while a dictation is in progress. On iPhone and iPad, the containing app keeps its audio session available while Keyboard Session is running so the system keyboard can start dictation. Audio buffers captured while Keyboard Session is idle are discarded. Apple Speech uses Apple’s system speech framework. Optional Parakeet and Whisper-family models run locally after you choose to download them.
Optional refinement
Refinement is off by default. If you enable Apple on-device refinement, Murmr sends the deterministic text result to the supported on-device Apple model. Murmr does not silently fall back to a cloud provider.
Keyboard Full Access
Full Access lets the Murmr system keyboard communicate with the Murmr app through the shared App Group. The containing app records and processes audio, then returns completed text to the keyboard. Audio and transcription remain on the device.
History, Dictionary, and Learning
Mac can retain up to 500 local dictations in History. Dictionary rules and opt-in Learning data are stored locally. iPhone and iPad keep no transcript history by default.
Clipboard handling
On Mac, Murmr temporarily uses the clipboard to insert accepted text into the active app, then restores the prior clipboard contents when the destination has not replaced them.
Downloads, deletion, and export
Optional model weights download only after you request them. Supported model, History, Dictionary, and Learning controls let you inspect, delete, or export the corresponding local data where the product offers those controls.
Company
Murmr Voice is provided by Murmr Labs. Product and privacy updates are published at murmrlabs.ai. Privacy questions can be sent to privacy@murmrlabs.ai.