Hold a hotkey
Use the configured shortcut or toggle mode. A clear menu-bar state tells you when VocaMac is listening.
Microphone + Input Monitoring permissions keep the control system-wide.
native macOS voice typing
Say it once.
Your Mac types.
VocaMac listens when you ask, transcribes on your Mac, and puts the words at your cursor. No account. Open source. Built for Apple Silicon.
macOS 14+
Apple Silicon
Beta · v0.9.0
AGPL-3.0
VocaMac
Ready to dictate
Hold Right Option to record
localVocaMac puts the words where I am already working
how it works
A small loop that stays out of the way.
VocaMac lives in the menu bar, so dictation starts where your work already is.
Say what you mean
Audio is captured in memory and passed to the speech-to-text model selected in Settings.
Choose a language hint when automatic detection needs help.
Keep working
When recording ends, the transcript is inserted at the cursor through macOS accessibility APIs.
Some secure fields and apps can restrict system text insertion.
choose your model
The model is a setting, not a mystery.
v0.9.0 includes WhisperKit with CoreML acceleration plus Parakeet, Apple Speech, and specialized ONNX models. VocaMac keeps model choice local, visible, and reversible.
First download
Models download from argmaxinc/whisperkit-coreml and are cached locally. After a model is available, transcription can run without sending audio to a cloud service.
More than Whisper
v0.9.0 brings the engines and models in the second table into the shipping release. Check the release notes for their language and macOS requirements.
| Model | Download | RAM | Character |
|---|---|---|---|
| Tiny | 39 MB | 1 GB | Fastest. Good for short commands and quick notes. |
| Base | 142 MB | 1.5 GB | A small step up in accuracy for very little extra weight. |
| Small | 466 MB | 2 GB | A sensible default on 8 GB Macs. |
| Large v3 Turbo (Compact) | 632 MB | 4 GB | The best accuracy-per-second trade-off for most people. |
| Distil Large v3 (Compact) | 594 MB | 4 GB | Distilled for speed. English-focused. |
| Distil Large v3 Turbo | 600 MB | 4 GB | The fastest of the distilled variants. |
| Large v3 Latest (Compact) | 626 MB | 5 GB | Full large-model accuracy, compressed. |
| Large v3 Latest | 3.1 GB | 10 GB | Highest accuracy. Wants a 16 GB Mac and patience. |
| Model | Engine | Download | Note |
|---|---|---|---|
| Parakeet v3 | Parakeet | 0.7 GB | Fastest engine. 25 European languages + Japanese, auto-detected. |
| Parakeet v2 | Parakeet | 1.2 GB | Fastest engine. English only, highest recall. |
| Parakeet 110M | Parakeet | 0.2 GB | English only, smaller download and faster first load. |
| Apple Speech | Apple Speech | Managed by macOS | macOS 26+ system engine (SpeechAnalyzer). Roughly 30 locales. |
| Moonshine v2 Tiny | sherpa-onnx | 60 MB | CPU-only. Very low-RAM Macs, English. |
| Moonshine v2 Base | sherpa-onnx | 190 MB | CPU-only. Low-RAM Macs, English. |
| SenseVoice | sherpa-onnx | 240 MB | CPU-only. Chinese, Japanese, Korean, Cantonese, English. |
| GigaAM v3 | sherpa-onnx | 270 MB | CPU-only. Russian, with punctuation. |
| Canary 180M Flash | sherpa-onnx | 320 MB | CPU-only. English, Spanish, German, French. |
real product states
A Mac app you can actually see working.
VocaMac is a menu-bar tool, not a dashboard pretending to be one. These are the states that matter while you dictate.
a precise privacy promise
Your voice has a path.
We show you which one.
VocaMac does not need a Voca cloud service to transcribe. The important boundary is the model running on your Mac.
- Microphone You hold the hotkey
- In-memory audio Held for the recording
- WhisperKit / CoreML Runs on your Mac
- Transcript at cursor Where you were typing
There is no extra stop. The path above has no Voca speech server in it. Dictation audio is not uploaded for transcription.
What stays local
During transcription, your audio and the selected model stay on the Mac. VocaMac has no Voca-hosted speech endpoint.
What uses a network
Initial model downloads, release downloads, and the app's update check use their documented sources. Apple may manage language assets for system speech.
What you can inspect
The application is open source under AGPL-3.0. Preferences, local model management, and the injection path are visible in the repository.
a focused tool, with real controls
Make the small details yours.
VocaMac keeps its surface compact while giving you control over the parts that change daily use.
Hotkeys and modes
Choose push-to-talk or double-tap toggle, then keep a shortcut that fits your hands.
Explore activationModels and language
Pick a local model and let it follow automatic language detection or a language hint.
Explore model choiceMicrophone and feedback
Choose the input device and use menu-bar, audio-level, or cursor feedback when you need it.
Explore feedbackYour vocabulary
Add names and terms that matter to you. Custom vocabulary is processed with the Whisper path.
Explore vocabularyBeta · v0.9.0
Pick your way in.
The signed release is ready for macOS 14+ and Apple Silicon. A first model download may need an internet connection.
Install from the tap
Use the cask when you want upgrades to fit into your normal Mac workflow.
$ brew tap vocahq/vocamac
$ brew trust vocahq/vocamac
$ brew install --cask vocamacUpdates: brew upgrade --cask vocamac
Download the signed app
Open the latest DMG, drag VocaMac to Applications, and launch it from Spotlight.
Download v0.9.0 DMGVocaMac-0.9.0-arm64.dmg · 99 MB · Developer ID signed and notarized by Apple
See releases and checksumsBuild from source
For contributors and people who prefer to build locally. Ad-hoc builds may require permissions again after a rebuild.
$ git clone https://github.com/VocaHQ/vocamac.git
$ cd vocamac
$ make install- System
- macOS 14 Sonoma or later
- Hardware
- Apple Silicon only. There is no supported Intel build.
- Permissions
- Microphone, Accessibility, Input Monitoring
- License
- GNU Affero General Public License v3.0
open source by default
A small Mac utility with nothing to hide.
Read the source, inspect the release, or help shape the next build. VocaMac is part of the wider Voca family of native tools.
Explore the Voca family questions, answered plainly
The useful bits before you install.
Does my voice leave my Mac?
During transcription, audio is passed to the selected local speech-to-text model on your Mac. VocaMac does not send dictation audio to a Voca cloud service. Model downloads and update checks are separate network actions.
Do I need internet access?
You need internet access to download the app or a model that is not on your Mac yet. After the model is available, on-device transcription can work offline. VocaMac may still check GitHub for application updates when that feature is enabled.
Which Macs can run VocaMac?
The current release supports macOS 14 Sonoma or later on Apple Silicon. Intel Macs are not supported by the released arm64 build.
Why does VocaMac need permissions?
Microphone access captures speech. Accessibility and Input Monitoring let the menu-bar app hear its configured shortcut and insert the resulting text into the app where you are working. Secure fields and individual apps can still restrict insertion.
Which model should I choose?
Start with Small on an 8 GB Mac, then choose a compact Large or Distil variant when accuracy matters and your Mac has the headroom. Tiny and Base are useful when a smaller download or lower memory use matters more.
What is available beyond Whisper?
v0.9.0 includes Parakeet (NVIDIA TDT models on the Apple Neural Engine), Apple Speech (the macOS 26+ system engine), and sherpa-onnx (CPU-only specialized models: Moonshine, SenseVoice, GigaAM v3, and Canary 180M Flash). See the model table above and the release notes for language and macOS requirements.
ready when you are
Let your voice stay close to the work.
Install VocaMac, choose a model, and put the words where your cursor already is.
macOS 14+Apple SiliconAGPL-3.0