What is voxtral-dictate?

Voxtral-dictate is a voice dictation tool for Linux that lets a person type by speaking, using AI transcription from Mistral's Voxtral model. It is built specifically for Ubuntu running the GNOME desktop on Wayland, and it works inside any application, whether that is a terminal, a text editor, or a web browser. The way it works is simple in practice: pressing a keyboard shortcut, Alt+T by default, starts recording from the microphone, and pressing it again stops the recording. The recorded audio is then sent to the Voxtral transcription service, the resulting text is copied to the clipboard, and a tool called ydotool simulates a paste keystroke so the transcript appears right at the cursor. If text was already selected when the shortcut was pressed, the transcript replaces it. A small overlay bubble, added through a GNOME Shell extension, shows a red dot while recording and a yellow dot while transcribing so the user can see the current state. Setting it up requires a Linux system running Ubuntu with GNOME on Wayland, a microphone, and an API key from Mistral for using the Voxtral transcription models. Installation is done through a setup script that installs several required packages, configures permissions needed for simulating keystrokes, asks for the Mistral API key, installs the GNOME Shell extension, and registers the chosen keyboard shortcut. A one time log out and back in is needed afterward for the new permissions and extension to take effect. The API key is stored securely in the GNOME keyring rather than as plain text, though a plaintext fallback exists in the config file if no keyring entry is set. Settings can be adjusted in a config file, including which Voxtral model to use, whether to force a specific spoken language, a maximum recording duration, and which paste keystroke combination to use, since some terminal and desktop apps expect different paste shortcuts.