ShellVoice

Changelog

What's new.

Every ShellVoice build, with what changed. Planned work is described on the developers page, not here.

  1. September 2026

    0.1.0

    In preparation

    The first ShellVoice build: dictation anywhere, on-device neural voices, and a command line that lets your tools speak.

    New
    • Global voice input: hold ⌥ Space (or toggle) in any app, speak, and the transcript is inserted where you were typing. Clipboard fallback when insertion isn't possible.
    • Local transcription with whisper.cpp (Metal-accelerated) using the tiny.en, base.en or small.en models, downloaded in-app. Optional OpenAI-compatible cloud provider.
    • Custom dictionary that fixes technical terms ("use query" → useQuery) and biases recognition toward your vocabulary.
    • Text to speech with free on-device neural voices (Piper): Nova, Miles, Ryan, Alba, Amy and Lessac, plus the native macOS voices. Play, pause, resume, stop, replay, speed and volume.
    • Floating recording overlay that appears over every app and Space, including fullscreen apps, without stealing focus.
    • Companion CLI: shellvoice speak, notify, status, stop and config, with stdin piping. Installable from Settings → Developer.
    • Spoken and system notifications for CLI messages, each with its own toggle.
    • Menu-bar integration, launch at startup, start minimized.
    • Local history of dictations and speech — optional, capped, clearable, stored only on your Mac.
    • Guided first-run setup: microphone, voice, shortcut, model download and a live dictation test.