Handy is a free, MIT-licensed dictation app with an unusual design goal: its author states the aim is to be the most forkable app in the category rather than the best one. Press a shortcut, speak, and the text appears in whatever field has focus — entirely offline, with no account and no payment. At 29,000 GitHub stars it's the most-starred project in this directory, and it is the only serious dictation app that treats Linux as a real platform rather than an afterthought.
At a glance
Platforms
macOS (Intel and Apple Silicon), Windows x64, Linux x64
Pricing
Free, MIT licence
Free tier
The whole app
Processing
On-device
Models
Whisper Small/Medium/Turbo/Large with GPU acceleration, Parakeet V3, custom GGML models
Offline
Yes, entirely
Integrations
Any text field; CLI parameters for remote control; Raycast extension
Linux actually works. Ubuntu 22.04 and 24.04 are supported out of the box with both X11 and Wayland handled. Every commercial competitor in this directory skips Linux entirely.
Free with no catch. No tier, no quota, no telemetry, no account. The author's framing is that accessibility tooling shouldn't sit behind a paywall.
Parakeet V3. CPU-optimised with automatic language detection, which makes it viable on machines without a usable GPU — the practical difference between Handy working and not working on older hardware.
Voice activity detection. Silero VAD trims silence so you're not transcribing dead air.
Genuinely extensible. MIT licence, 2,600 forks, and a stated design goal of forkability. If you want to build on top of a dictation app, start here.
Where it falls short
Wayland is rough at the edges. Text input needs wtype or dotool, the recording overlay is disabled by default because some compositors break on it, and certain GPU driver combinations need WEBKIT_DISABLE_DMABUF_RENDERER=1 before the UI renders. It works, but you may be reading issue threads.
Noticeable delay. Two to five seconds after you stop speaking, depending on model and hardware. This is a transcribe-on-release app, not a streaming one.
Two recurring bugs. The first word of a transcription is sometimes clipped, and auto-paste can land text in the wrong window if you switch apps before processing finishes. Both are documented in the tracker.
No cleanup pass. You get a transcript, not edited prose. There's no LLM rewriting filler words into a finished draft the way the paid tools do.
Accuracy tracks your hardware. Whisper Small on an old laptop is meaningfully worse than what the reviews describe running Large.
Pricing
Free. MIT licence, no paid tier, no hosted component to pay for.
Who it's for
Linux users, first and foremost — there is essentially no competition. Beyond that, anyone who wants offline dictation without a subscription and can live with a couple of seconds of latency and a raw transcript. If you want the text to arrive already edited, the paid cloud tools do that and Handy doesn't. If you're on a Mac and want a more finished product for a one-off fee, VoiceInk is worth the $29.
Last verified: 9 August 2026.
Similar to Handy
Handy: The free offline dictation app that treats Linux as a first-class target – Dictation Database
