Skip to content

YazSes vs Handy

Short answer: Handy is the better choice if you want dictation and nothing else, with the smoothest install and a large, active community behind it. YazSes is the better choice if you want voice commands, offline transcription of existing recordings, meeting capture with speaker labels, or Wayland without extra setup — or if accessibility features are the reason you are looking.

Both are free, open source, and run entirely offline on Linux, macOS and Windows. YazSes on macOS and Windows is in preview: supported, not yet verified on hardware by a person on a current build (see the validation matrix). Handy is by far the more popular project. This page is about which fits your job, not which has more stars.

At a glance

Handy YazSes
Licence MIT Apache-2.0
Built with Tauri (Rust) + React Python
Models Whisper (GPU when available), Parakeet V3 faster-whisper (CPU int8); Parakeet optional
Activation Hotkey, or push-to-talk Hold-to-talk
Linux Wayland Limited — needs wtype or dotool Runtime backend probe picks a working injector; not yet verified on GNOME and KDE hardware (#643)
Voice commands No Yes (regex grammar → key sequences)
Transcribe existing files No Yes (yazses transcribe)
Meeting capture + speaker labels No Yes (yazses meeting)
macOS / Windows Yes Yes (preview)

Where Handy is the better tool

Being straight about this matters, because for a lot of people Handy is the right answer.

  • It is more polished to install. A Tauri desktop app with a real installer is a friendlier front door than a Python daemon plus a permissions checklist. YazSes ships yazses doctor and yazses verify precisely because its install has more ways to go wrong.
  • It uses your GPU. Handy runs Whisper with GPU acceleration where available. YazSes deliberately targets CPU int8, so a strong GPU sits idle.
  • The community is much larger. More users means more bug reports, faster fixes, and more chance your exact hardware has already been hit and solved.
  • It does one thing. If you only want to dictate, everything below is weight you do not need.

Where YazSes does something Handy does not

  • Voice commands. Spoken phrases mapped by a fast regex grammar to real key sequences — "undo that", "save file", "go to line 42". Handy is dictation only.
  • Wayland without extra setup. Handy documents limited Wayland support requiring external tools. YazSes probes the session at runtime and selects a working injection backend — ydotool, wtype, xdotool, or clipboard paste — and yazses doctor reports which one it picked. A hardware-verified record on GNOME and KDE Wayland is still to come (#643).
  • Existing recordings. yazses transcribe <file> turns any audio or video file into text, Markdown, SRT, WebVTT or JSON, optionally with speaker diarization.
  • Whole meetings. yazses meeting captures hands-free and produces a speaker-labelled transcript, optionally with minutes written by a local LLM.
  • Accessibility. VAD calibration, a dysfluency-friendly mode for stuttered or dysarthric speech, and an activation layer where the hotkey is a replaceable part — an experimental EMG muscle sensor (off by default, no real-device evidence yet) is one such alternative.
  • An opt-in on-device learning loop. yazses tune proposes accuracy fixes from your own corrections, all encrypted and local.

The honest summary

If your answer to "what do you want it to do?" is "type what I say", Handy is an excellent choice and probably the easier one to live with.

If your answer includes "…and run commands", "…and handle my recordings", "…and take my meeting notes", or "…and work for someone who cannot hold a key", those are the jobs YazSes was built for, and they are not on Handy's roadmap as far as its documentation describes.

Trying both

They can coexist — just do not bind them to the same key.

bash <(curl -fsSL https://raw.githubusercontent.com/MSKazemi/yazses/main/install.sh)
yazses doctor
yazses enroll
yazses start