ShellVoice

Features

Everything in ShellVoice.

Two directions of voice — in and out — plus the details that make them feel native on a Mac. All of this is in the current build.

Dictation

Voice to text, wherever your cursor is.

The Dictation screen shows the live state of the pipeline — idle, listening, transcribing, inserted — and keeps your most recent transcript in view. The same field responds inside the overlay and on the home screen.

Global shortcut
⌥ Space by default; choose your own. If a combination is taken, ShellVoice tells you.
Hold or toggle
Push-to-talk feels like a walkie-talkie. Toggle mode starts and stops with a single press.
Automatic insertion
Text is pasted into the app that was frontmost when you pressed the key — even if ShellVoice is visible.
Three local models
tiny.en (75 MB), base.en (142 MB) or small.en (466 MB). Pick the accuracy your Mac can afford.
ShellVoice Dictation screen while listening: the voice field reacts to speech, the label reads “Listening — speak naturally”, with a Cancel button.

Recording overlay

Small, floating, out of the way.

A single pill with the voice field and one status word. It shows over fullscreen apps and every Space, never takes focus, and disappears when you're done.

  1. The floating ShellVoice overlay pill while listening: a small reacting voice field beside the words “Listening — Release to transcribe”.

    1Hold the shortcut and speak

  2. The floating ShellVoice overlay pill while transcribing: the voice field pulses beside the word “Transcribing”.

    2Release. Transcription runs locally

  3. The floating ShellVoice overlay pill after insertion: a calm ring beside the word “Inserted”.

    3Text lands where you were typing

Default shortcut ⌥ Space · hold to talk, or switch to toggle in Settings

Try it

See the listening state.

The same organic field that runs in the app, here in your browser.

Idle

In the app: hold ⌥ Space anywhere, or click the field

Demonstration. The visual reacts to simulated audio, or to your microphone level if you enable it. Levels are measured in your browser only — nothing is recorded, transcribed, or sent anywhere. Real transcription happens in the desktop app.

Text to speech

Voices worth listening to.

Type or paste, press ⌘ ↩, and listen. The same engine reads CLI notifications aloud, so “Build finished” sounds like a person, not a robot.

Neural voices, on-device
Nova, Miles, Lessac, Ryan, Amy and Alba — free Piper voices you download once and keep.
macOS voices
Every system voice installed on your Mac appears in the same list.
Play, pause, resume, stop
Real playback controls, plus replay for the last thing spoken.
Speed and volume
0.5× to 2× speed, independent volume, previewed before you commit.
ShellVoice Text to Speech screen: a text box containing “Build completed. 142 tests passed. The preview deployment is live.”, a Play button, and Voice, Speed and Volume controls with the Nova voice selected.

Hear them

The voices, straight from the app.

This list is imported from the app's voice catalog, and each preview was synthesized with the app's own engine.

  • NovaRecommended

    warm US female · en-US · 61 MB · on-device

  • Miles

    calm US male · en-US · 61 MB · on-device

  • Lessac

    clear US female · en-US · 61 MB · on-device

  • Ryan

    expressive US male · en-US · 115 MB · on-device

  • Amy

    friendly US female · en-US · 61 MB · on-device

  • Alba

    Scottish female · en-GB · 61 MB · on-device

Each clip is “Build completed. 142 tests passed. The preview deployment is live.”, synthesized with the app's Piper engine and the same voice files it downloads. Plus every macOS system voice already on your Mac.

Custom dictionary

It learns how you say things.

Speech recognizers don't know your codebase. The dictionary maps what the recognizer hears to what you meant — “use query” to useQuery, “tan stack” to TanStack — and feeds those terms back to the model so the next dictation gets them right the first time.

ShellVoice Settings scrolled to the Transcription and Custom dictionary sections: a local Whisper model picker and dictionary entries such as “use query → useQuery” and “tan stack → TanStack”.

And the rest

The details that make it feel native.

Menu bar app

Lives in the menu bar. Launch at login, start minimized, and open any screen from the tray.

Spoken notifications

Messages from the CLI are read aloud and, if you like, shown in Notification Center. Each has its own switch.

Local history

Recent dictations, spoken text and notifications, with copy and clear. Optional, capped, and only on your Mac.

Guided first run

Microphone, voice, shortcut, model download and a live test — the setup takes a minute and ends with your first dictation.

Cloud transcription, optional

Point ShellVoice at any OpenAI-compatible endpoint if you need to. Local stays the default.
Planned

On the roadmap

Agent hooks for Claude Code, Codex and Gemini CLI, streaming transcription, voice commands, per-app profiles, Windows.

History

Everything you said, only where you can see it.

Stored in a file readable only by your user account, capped, and gone the moment you clear it.

ShellVoice History screen listing recent dictations, spoken text and CLI notifications with timestamps, and a Clear button.
ShellVoice first-run welcome screen: the voice field above the ShellVoice wordmark, the line “Your terminal, now speaking.”, and a Get Started button.

Give your Mac a voice.

Dictate anywhere. Hear your tools. Keep everything on your machine.

macOS
Available for macOS
Windows
Windows planned
Download ShellVoice

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