Hold Right ⌥ and talk.
When you let go, your words are pasted where your cursor is, already cleaned up.
peluni isn't released yet, so there's nothing to download. With Xcode 16.3 on a Mac with Apple Silicon, you can build it from the source today. To hear when it's out, watch the repository on GitHub and pick Releases.
Reminders
Errands
- Return the library books
- Pick up the charger and oat milk on the way home.Heard:
umpick up theuhcharger andlikeoat milk on the way home
Cleanup levels
Whisper writes down everything you say, false starts included. You choose how much of it gets pasted. If cleanup fails or takes more than ten seconds, you get the raw transcript.
None
raw transcript
um so the demo moved to thursday no wait friday and uh we still need the slides the script a backup laptop and the hdmi adapter
Lightdefault
fillers and punctuation
So the demo moved to Thursday, no wait, Friday, and we still need the slides, the script, a backup laptop and the HDMI adapter.
Medium
also grammar and false starts
The demo moved to Friday, and we still need the slides, the script, a backup laptop and the HDMI adapter.
High
also structure and lists
The demo moved to Friday. We still need:
- The slides
- The script
- A backup laptop
- The HDMI adapter
Vocabulary
Vocabulary
- Priya
- pelunisounds like: pell oony
- MLXsounds like: em el ex
- Kubernetessounds like: cooper netties
Heard: ship the pell oony build to priya
Ship the peluni build to Priya.
Add the words whisper keeps getting wrong, like your colleagues' names or the acronyms your team uses. peluni passes them to whisper as hints, and any alias you list under “sounds like” is replaced the same way every time.
Where your audio goes
Turn Wi-Fi off and keep dictating.
Wi-FiOff
Transcription and cleanup run in memory on your Mac, and the app only goes online to fetch a model when you ask for one. It never types into a password field. The code for all of this is on GitHub.
Building it yourself
Free and open source. Version 0.0.1 is an early development build, not ready for a beta yet. Once built, it runs on macOS 14 (Sonoma) or later with Apple Silicon (M1 or later).
- Install Xcode 16.3 or later, which itself needs macOS 15.2 on a Mac with Apple Silicon. On Xcode 26, also run
xcodebuild -downloadComponent MetalToolchainonce. - Clone the peluni repository from GitHub.
- In its folder, run
make run. The first build takes several minutes; later ones are quicker. - Allow Microphone and Accessibility when the setup window asks, then download the Base speech model it offers.
With the base model, the text typically lands about a second after you let go. Double-tap Right ⌥ to keep recording with your hands free, and press it once more to stop. Esc cancels at any point, and a tap shorter than 0.3 seconds is ignored.
Build requirements and speech model sizes are in the README. Anything else: open an issue on GitHub.