Handy

Turn a shortcut into local recording, VAD, transcription, and text insertion

Handy supports push-to-talk, toggle recording, and a combined hold-or-toggle interaction. When recording stops, voice activity detection removes silence before the audio is passed to the local transcription model. The resulting text is then inserted into the application the user was already working in instead of remaining inside a dedicated editor. That makes Handy useful as an operating-system-level dictation layer even for applications that provide no native speech input.

Sources: [1]

Choose Whisper acceleration or CPU-oriented Parakeet V3

The transcription engine supports several Whisper sizes, including Small, Medium, Turbo, and Large, with GPU acceleration where available. Parakeet V3 provides a CPU-oriented alternative with automatic language detection; the README reports performance around five times real time on a mid-range CPU in its example hardware. This gives users a practical choice among accuracy, model size, inference speed, and available GPU resources while keeping the primary speech-recognition path on the local machine.

Sources: [1]

Integrate dictation into launchers and automation with CLI and Raycast

Handy's settings interface is written in React and TypeScript, while Rust handles audio I/O, keyboard events, ML inference, and operating-system integration through Tauri. The CLI can toggle transcription, start a recording with post-processing, cancel an operation, or launch the app hidden, and a Raycast extension exposes actions for recording, transcript history, dictionary management, models, and languages. This makes the app easier to embed into custom keyboard and launcher workflows instead of requiring all interaction through its own window.

Sources: [1]

For private dictation across many desktop applications

Handy fits users who regularly write across browsers, email, chat, IDEs, and note-taking tools and want one consistent speech-input workflow rather than app-specific dictation features. Its local transcription architecture is particularly relevant when raw voice should remain on the user's machine. CLI controls also make it attractive to users who want to integrate transcription into desktop-environment shortcuts, launchers, or window-manager automation.

Sources: [1]

Core STT is local; optional post-processing and OS integration vary

The speech-to-text pipeline itself runs locally, but Handy also has optional transcript post-processing with configurable providers. Its settings include OpenAI-style API-backed providers, custom endpoints, and an Apple Intelligence provider. When an external provider is selected, the text being post-processed may leave the local machine, so the product's offline privacy claim should be understood primarily in relation to its core transcription path rather than every optional workflow.

Operating-system integration also varies. macOS requires microphone and accessibility permissions and Bluetooth microphone use can temporarily change audio quality. On Linux Wayland, reliable text insertion may require tools such as wtype, dotool, or in some environments ydotool, and global shortcuts may need to be owned by the desktop environment or window manager. The README also documents configuration-dependent Whisper crashes on some Windows and Linux systems. Testing the intended model, microphone, and text-insertion path on the target machine is therefore important. Handy is licensed under the MIT License.

Sources: [1][2][3]