ShellVoice

Download

ShellVoice for macOS

Version 0.1.0 is complete and being prepared as a signed, notarized build. Until it's published, you can build it from source.

Build from source

Version 0.1.0 for macOS 12.0+ · Signed builds are being prepared

This release

What you're getting.

Version
0.1.0
Status
Signed builds in preparation
macOS
12.0 or later · Apple Silicon and Intel
Windows
Planned

System requirements

Runs on any Mac from the last few years.

macOS 12 or later

Monterey through the current release, on Apple Silicon or Intel.

A microphone

Built-in or external. Choose it in Settings → Dictation.

Disk space for models

Whisper models are 75–466 MB; each neural voice is about 61–115 MB. Downloaded once, in-app.

Build from source

Build it yourself today.

ShellVoice is a Tauri 2 app with a Rust backend. Building takes a few minutes the first time because whisper.cpp compiles from source.

You need Rust (stable, via rustup), Node 20+, pnpm and cmake (brew install cmake).

The source repository link will appear here once it is public. The commands on the right are the ones it will use.

Local builds are ad-hoc signed, so macOS forgets the Accessibility grant on every rebuild. See troubleshooting below.

Terminal
pnpm installpnpm tauri build          # produces ShellVoice.app# the command-line companioncargo build --release -p shellvoice-cli# → target/release/shellvoice

Permissions

Two permissions, both explained.

ShellVoice asks only for what dictation needs. Nothing else is requested.

PermissionWhyRequired?
MicrophoneRecording your dictation. Audio is transcribed on-device and discarded.Yes, for dictation
AccessibilitySynthesizing ⌘ V to insert the transcript into the app you were using.For automatic insertion. Without it, transcripts go to the clipboard.

To grant Accessibility: open ShellVoice, use the banner's “Open Accessibility settings” button (or System Settings → Privacy & Security → Accessibility), enable ShellVoice, then click “Re-check”.

First run

A minute of setup.

The guided setup runs once and ends with a live dictation test.

  1. 1

    Microphone

    Grant access and watch the field react to your voice.

  2. 2

    Voice

    Pick a neural voice — Nova is recommended — or a system voice, and hear a preview.

  3. 3

    Shortcut

    Keep ⌥ Space or record your own. Conflicts are reported immediately.

  4. 4

    Model

    Download base.en (142 MB) or choose tiny or small.

  5. 5

    Test

    Hold the shortcut, say “Hello ShellVoice”, and see it land in the test field.

  6. 6

    Done

    ShellVoice moves to the menu bar and waits for your shortcut.

Troubleshooting

If something doesn't work.