Most dictation tools that work this well are sending your voice to someone else's servers. Handy inverts the usual tradeoff: the audio never leaves your machine, yet it still does push-to-talk transcription that pastes straight into the active window. The more revealing detail is its stated goal — it explicitly does not aim to be the best speech-to-text app, but the most forkable one.
What Sets It Apart
- Fully local pipeline: speech is captured, run through voice-activity detection (Silero), and transcribed on-device, so dictation works on a plane or behind a firewall with no API keys or quotas.
- Model choice that matches your hardware: Whisper variants (Small / Medium / Turbo / Large) use the GPU when present, while Parakeet V3 is CPU-optimized and auto-detects language — a practical fallback for machines without a capable GPU.
- A Rust/Tauri core (whisper-rs, transcribe-rs, cpal, vad-rs, rdev) rather than an Electron wrapper, which keeps the binary small and the system-level shortcut and audio handling close to the metal.
Built To Be Forked
The "most forkable" framing is the real design decision here. The codebase is structured as accessibility tooling meant to be copied, modified, and rebuilt for niche needs — a custom shortcut scheme, a different model, a specialized workflow — rather than locked behind a polished but rigid product. That makes it as much a starting point as a finished app.
Great Fit If / Look Elsewhere If
Great fit if you want private, keyboard-driven dictation across Windows, macOS, or Linux, or a clean base to build your own voice tool on. Look elsewhere if you need cloud-grade accuracy on weak hardware, real-time streaming captions, or a turnkey product with vendor support — the local models trade some accuracy and polish for privacy and hackability.
