handy

Handy is a cross-platform desktop application that provides simple, privacy-focused speech transcription. It is free, open source, and extensible, designed to work completely offline without sending any audio data to the cloud. The core workflow is straightforward: press a configurable keyboard shortcut, speak, and release to have your words transcribed and pasted directly into whatever text field you are using. The application supports Windows, macOS (both Intel and Apple Silicon), and Linux. It uses local speech recognition models, primarily Whisper-family models (Small, Medium, Turbo, Large) with GPU acceleration when available, and Parakeet V3, a CPU-optimized model with automatic language detection. Silence is filtered using Voice Activity Detection (VAD) with Silero before transcription. Handy is built as a Tauri application combining a React + TypeScript frontend with Tailwind CSS for the settings UI, and a Rust backend for system integration, audio processing, and ML inference. Core libraries include transcribe-cpp for Whisper models, transcribe-rs for Parakeet models, cpal for cross-platform audio I/O, vad-rs for voice activity detection, rdev for global keyboard shortcuts, and rubato for audio resampling. The app offers multiple recording modes: hold-to-record, tap-to-toggle, hold-only, and toggle-only. It includes a debug mode accessible via keyboard shortcut, CLI parameters for remote control and startup customization, and a Raycast integration for controlling recording, browsing transcript history, managing dictionaries, and switching models and languages. For Linux users, Handy requires additional tools for text input depending on the display server: xdotool for X11, wtype or dotool for Wayland. The recording overlay is disabled by default on Linux to avoid focus-stealing issues. Global keyboard shortcuts on Wayland must be configured through the desktop environment or window manager, with CLI flags available for integration. Handy supports manual model installation for users behind proxies or in restricted network environments, and can auto-discover custom Whisper GGML models placed in the models directory. Release artifacts are signed with Tauri's updater signature format and can be verified using minisign. The project is actively developed with ongoing work on debug logging, macOS keyboard improvements, opt-in analytics, settings refactoring, and Tauri command cleanup. Known limitations include Bluetooth headset microphone issues on macOS, fn/Globe key shortcuts only working on Apple keyboards, Whisper model crashes on certain system configurations, and limited Wayland support.