The speech-to-text app built to be forked

What it does

Handy is a Tauri-based desktop app that listens to your microphone when you hit a global shortcut, runs your speech through a local model, and pastes the resulting text into whatever field currently has focus. It keeps everything on your machine—audio, inference, and output—using either OpenAI Whisper variants (with GPU acceleration where available) or the CPU-optimized Parakeet V3 model. The result is privacy-first voice typing that never phones home.

The interesting bit

The project openly admits it is “trying to be the most forkable one,” not the best speech-to-text app. That unusual humility is backed by a Rust backend wired for system integration—global hotkeys, audio I/O, voice-activity detection, and model swapping—while exposing a React settings UI. It is essentially a well-organized glue layer around whisper-rs, transcribe-rs, and cpal, which is exactly what makes it extensible.

Key highlights

• Runs fully offline with a choice of Whisper or Parakeet V3 models; Parakeet claims ~5× real-time speed on mid-range Intel hardware and automatic language detection.
• Cross-platform support for macOS (Intel and Apple Silicon), x64 Windows, and x64 Linux.
• Provides global keyboard shortcuts, push-to-talk, CLI remote-control flags, and Unix-signal hooks for integration with window managers.
• Includes a Raycast extension for browsing history, managing dictionaries, and switching models without opening the main window.
• Uses Silero voice-activity detection to strip silence before sending audio to the transcription engine.

Caveats

• Whisper models crash on certain Windows and Linux configurations; the maintainers note the issue is configuration-dependent and are soliciting debug logs.
• Linux support is uneven: Wayland users must wire up external typing tools such as wtype or dotool, global shortcuts must be handled by the desktop environment, and the recording overlay is disabled by default because some compositors steal focus from the target window.
• The app is currently x64-only on Windows and Linux; macOS is the only platform with Apple Silicon support.

Verdict

Developers who want a hackable, offline dictation layer they can customize or embed into their own workflow should take a look. If you need a turnkey, universally stable experience across all Linux setups—or an ARM64 Windows build—this is not there yet.