Press Alt+Space, speak, press it again. Groq Whisper transcribes, Gemini cleans up the grammar, and the text lands wherever your cursor already is.
Nothing else leaves the machine. A small background daemon, xhispertoold, holds the virtual keyboard so the text can land in any window.
Starts recording. The status (recording...) types itself at your cursor.
Groq Whisper transcribes in real time once you stop.
Gemini 3.1 Flash Lite fixes grammar and filler words without rephrasing your sentence.
Result lands via a virtual uinput keyboard. Silence cancels automatically, nothing gets pasted.
The polish prompt is deliberately narrow: fix obvious errors, do not restructure sentences, do not add tone.
um so basically the the deploy script needs to check for uh the env file before it runs right so we dont crash on missing keys
So the deploy script needs to check for the env file before it runs, so we don't crash on missing keys.
No accounts, no telemetry. Two free-tier API keys and it runs entirely on your own machine.
Works in browser, terminal, IDE, chat, email, any text field, any window. No browser extension, no accessibility API integration required.
Audio stays on your machine. Only the audio blob leaves, for transcription and polishing, nothing else.
Every dictation saved with a timestamp, raw and polished versions, re-copyable from a GUI popup or the terminal.
GPL 3.0 for the wrapper, MIT for the xhisper core it's built on.
Groq Whisper and Gemini both ship free tiers wide enough for daily dictation. No credit card required to start.
| Service | What it does | Free key |
|---|---|---|
| Groq | Speech-to-text (Whisper) | console.groq.com/keys |
| Gemini | Text polishing (Gemini 3.1 Flash Lite) | aistudio.google.com/apikey |