Local first.
By your choice.

Local dictation needs no provider account or API key. Model downloads and update checks use the network. Murmur has no transcript or audio telemetry.

Transcript history has no cloud sync. Retention defaults to 30 days, with six-month and forever options. Turning History off keeps new transcripts only until the app quits; it does not erase old history or disable recovery audio and learning.

Cloud transcription and cleanup are configured separately. Settings shows the destination and data disclosure. API keys are stored in macOS Keychain.

Optional screen context is off by default. Enabling it for AI cleanup can send selected text, clipboard text, and focused-window text alongside the transcript to your chosen route.

Failed dictation audio, unfinished meeting chunks, and call audio waiting for speaker separation can remain locally for up to 24 hours. Expiry runs while Murmur is running, not while it is quit.

Correction learning is on by default. It observes edits to the inserted span for up to a minute and saves learned spellings separately from history. Disable learning or require readable fields in Settings → Dictionary.

Meetings start after your action. Check the audio source and notify participants before recording. Cloud transcription sends meeting audio; cloud notes send transcript content and your notes. These routes are separate choices.

Settings → History & Privacy → Delete All Data removes local dictation and meeting history, meeting notes, retry audio, dictionary and learning data, held transcripts, and Murmur’s API keys.

Settings and downloaded models remain. This cannot remove copies in cloud services, backups, or other applications.