Dictation that lives on one key.

Fast local-first dictation from a configurable Globe/Fn shortcut.

Press it, say what you mean, and the text is typed into whatever you were already working in.

Download for Apple Silicon

Version 0.4.3 (build 24). Requires macOS 26+ on Apple Silicon.

Developer ID signed and Apple notarized, so it opens normally the first time. See the install steps.

The recording overlay, at its real proportions: a live waveform and the elapsed time.

What Murmur does

  • One key, then it types

    Press the Globe or Fn key, say what you mean, and the text is typed into whatever app you were already working in.

  • Press to toggle, or hold to talk

    Recording can start on one press and finish on the next, or run only while the key is held down. The trigger key and the activation style are both settings.

  • Transcription stays on this Mac

    After the one-time model download, speech is transcribed on-device. There is no account to create and nothing to sign in to.

First run and privacy

On first use the app downloads about 470 MB of Parakeet V3 model files directly from FluidInference on Hugging Face. Those files are not inside the DMG and are not hosted on this site. The download goes straight from Hugging Face to your Mac.

Once the model is on disk, transcription runs on-device. There is no account to create and nothing to sign in to.

Give the first download a good connection and some patience. Later launches skip it entirely.

Installing

  1. Download the disk image

    Use the Download button at the top of this page. It always points at the newest build.

  2. Drag Murmur.app to Applications

    Open the DMG and drag Murmur.app across, then eject the disk image. Updates arrive inside the app after that.

  3. Open Murmur from Applications

    Open Murmur.app normally from Applications. It is Developer ID signed and Apple notarized, so a plain double-click is all it takes.

  4. Grant Microphone and Accessibility

    macOS asks for both on first use. Microphone captures what you say; Accessibility is what lets the app type the result into the app you were already in.

  5. Wait for the model to download

    About 470 MB, once. The app shows the progress, and dictation is ready when it finishes.

Signing and updates

Developer ID signed and Apple notarized

Apple has verified this build, and the notarization ticket is attached to the app and the disk image. Open it normally from Applications after dragging it out of the DMG.

The app ships with a Sparkle update feed, and every update archive is cryptographically signed, so new builds can be found and verified. Sparkle installs the same Developer ID signed and notarized app carried by the DMG. You can read the feed at appcast.xml if you want to see what the app sees.

Release details

Version
0.4.3 (build 24)
Minimum macOS
26.0 or later
Architecture
Apple Silicon (arm64)
Code signing
Developer ID Application, Apple notarized
Disk image
Murmur-0.4.3.dmg
Disk image SHA-256
3b9be21438b92342aaf7ebdf254c1d3ff2857d0877e9f19f101e762fd85841e6
Update archive
Murmur-0.4.3.zip
Archive SHA-256
205f9a715232f667687bd718b4b13bf2418b8326e1ae703d24e1983e05442a17
Speech model
FluidInference/parakeet-tdt-0.6b-v3-coreml
Model revision
aed02740059203c4a87495924f685de3722ae9ce
Model bundled
No, downloaded on first run

Verify a download before you open it:

shasum -a 256 Murmur-0.4.3.dmg

The same values are published as machine-readable JSON at release-metadata.json.