
VOXD is a Linux dictation tool that adds the piece most of its neighbours are missing: an AI post-processing pass. Transcription runs on whisper.cpp locally, and the cleanup that turns a transcript into finished text can run through llama.cpp on your own machine or, if you'd rather, through OpenAI, Anthropic, or xAI. It types straight into whatever window has focus, offers CLI, GUI, tray, and voice-triggered interfaces, and explicitly doesn't need a GPU.
| Platforms | Arch, Ubuntu, Fedora, Pop!_OS, Mint, openSUSE. X11 and Wayland |
| Pricing | Free, MIT licence |
| Transcription | whisper.cpp, locally |
| Post-processing | llama.cpp locally, or OpenAI / Anthropic / xAI |
| Languages | 99+ |
| GPU | Not required |
| Interfaces | CLI, GUI, system tray, voice-triggered mode |
| Maturity | ~288 stars, 254 commits, actively developed |
Free, MIT. Cloud post-processing costs whatever the provider charges.
Linux users who want the AI-polished output of the commercial tools and are willing to run two local models to get it — or who'll accept a cloud API for the cleanup step. It's the most capable Linux option on paper. If you want something more proven and are happy with a raw transcript, Vocalinux or Handy are safer picks.
Last verified: 9 August 2026. No independent reviews located; details are from the project's own repository.