cjpais/handy

Handy is a free, open source desktop application that turns spoken words into typed text, running entirely on your own computer. You press a configurable keyboard shortcut, speak, release, and the words you said appear in whatever text field you were focused on. Because the work is done locally, your voice never leaves your machine, which is the project's main pitch: a privacy-focused, offline speech-to-text tool that anyone can use, fork, or extend. Under the hood, Handy is built as a Tauri application, meaning the settings interface is a web frontend written in React and TypeScript with Tailwind CSS, while the heavy lifting (audio capture, system integration, and the actual speech recognition) is done in Rust. When you talk, Handy uses Voice Activity Detection through a model called Silero to filter silence, then transcribes the audio with one of two model families: OpenAI's Whisper models (Small, Medium, Turbo, or Large, with GPU acceleration when available) or Parakeet V3, a CPU-friendly model that detects language automatically. Audio plumbing, global keyboard shortcuts, and resampling are handled by smaller Rust libraries like cpal, rdev, and rubato. There is also a command-line interface for toggling recording, cancelling, or launching the app hidden, which makes Handy convenient to wire into automation scripts or tools like Raycast. It is cross-platform, with builds for Windows, macOS, and Linux, and is installable through Homebrew or winget as well as direct downloads. The natural users are people who want quick dictation in any app without depending on a cloud service: writers, developers, accessibility users, and anyone wary of sending audio to external servers.