On-device dictation for macOS, in the style of Wispr Flow. Press Option-Space, speak, press it again, and the text is pasted wherever the cursor is, in browsers and native apps alike. Nothing leaves the Mac: speech-to-text is Apple's SpeechTranscriber and the optional cleanup is Apple Intelligence (Foundation Models).
Requires macOS 26 or later on Apple silicon; cleanup also needs Apple Intelligence turned on.
./scripts/build.shThis builds the app, draws its icon and installs it at /Applications/Murmur.app. On first use, allow Microphone and Accessibility (Accessibility lets it paste for you; without it the text is left on the clipboard).
- Option-Space starts listening; a pill at the bottom of the screen shows your voice level. Option-Space again stops and pastes.
- Menu bar: Clean Up with Apple Intelligence (on or off), Edit Cleanup Rules…, Open at Login.
- Rules file (
~/Library/Application Support/Murmur/rules.md): plain-English rules for the cleanup, then a## Replacementslist ofheard → writtenswaps. The swaps run as find-and-replace, even with cleanup off, because the on-device model applied a word list unreliably.
swift build -c release
.build/release/Murmur --transcribe sample.aiff # prints the raw and cleaned transcript
say -o sample.aiff "um so I I think we should meet at three, no, at four" # makes a sampleThe build is ad-hoc signed, so macOS forgets the Accessibility grant after each rebuild: tick Murmur again in System Settings › Privacy & Security › Accessibility.