Handy is a free, 🏷️#open-source speech-to-text app that transcribes your voice into any text field on your computer. Created by CJ Pais, it runs entirely on-device — hold a keyboard shortcut (push-to-talk) or toggle it on, speak, and the transcription is pasted into the active text box, with no audio ever sent to the cloud. Its design philosophy is deliberately narrow: one tool, one job, with accessibility tooling kept free rather than behind a paywall.
Handy's distinguishing trait is model choice — it bundles eight local speech-recognition models and lets you pick the trade-off between size, speed, and language coverage, from OpenAI's 📝Whisper (99+ languages) and NVIDIA's Canary to the lightweight Moonshine and the East-Asian-focused 📝SenseVoice. Whisper models get GPU acceleration via Metal on macOS and Vulkan on Windows and Linux, while the rest run CPU-only. The project is sustained by sponsors including Wordcab, Epicenter, and Bolt AI. It targets anyone who wants fast, private dictation — writers, developers, and accessibility users — without a subscription or an internet connection.
Key Features
- On-device transcription — All speech-to-text runs locally, so your voice never leaves the machine and dictation keeps working with no internet connection.
- Push-to-talk or toggle — Hold a shortcut while you speak, or tap once to start and stop, and the text drops straight into whatever field has focus.
- Swappable models — Choose among eight local models to balance footprint, speed, and language coverage, from tiny English-only Moonshine to multilingual Whisper.
- GPU acceleration — Whisper models use Metal on macOS and Vulkan on Windows and Linux for faster transcription; the other models run CPU-only.
- Free and open-source — No paywall and no account, with a public codebase the community can audit, extend, and contribute back to.