1. Use Copilot CLI’s native voice input first
Install or update GitHub Copilot CLI, start an interactive session with copilot, and authenticate if required. GitHub says the CLI is available with all Copilot plans; organization-provided access also depends on the organization’s Copilot CLI policy.
Run /voice. On first use, choose Continue to download the voice runtime, then download and activate the English or Spanish speech-to-text model. The runtime download can continue in the background while you use the CLI.
- Start an interactive copilot session.
- Run /voice and download the voice runtime.
- Download and activate the intended English or Spanish model.
2. Hold Space for short prompts
Hold the Space bar until recording begins, speak the prompt, and release Space. Copilot CLI transcribes the audio and inserts the recognized text at the cursor in the prompt area.
The text is not submitted merely because transcription finished. Read it first, especially when names, paths, commands, negations, or permissions could change what the coding agent does.
3. Toggle recording for a longer requirement
Press Control-X followed by V to start a longer recording. Speak the requirement, then press any key to stop and insert the transcription.
GitHub also documents /refine for removing filler, applying self-corrections, fixing grammar, and grouping related points. Treat local speech recognition and subsequent Copilot processing as separate boundaries: GitHub’s local claim covers transcription, not every action taken after the text enters the agent prompt.
4. Switch the microphone or language deliberately
Run /voice devices to choose another available microphone. Run /voice models to download or activate a different speech model.
GitHub currently documents English and Spanish dictation only, with English as the default. The active model and voice-enabled state are stored in Copilot CLI settings, typically under ~/.copilot/settings.json.
5. Add IraVoice only for the cross-app or code-name workflow
Use Copilot CLI voice input when you are already at its prompt: it is native, local for transcription, and gives you editable text before submission. No separate dictation app is needed for that job.
Use IraVoice when the same held-key habit should extend to email, notes, browser, editor, and terminal fields on Mac, Windows, and Linux, or when file and function names keep coming out wrong. In a terminal working inside a Git project, IraVoice spells that project’s code names as the code does—“use effect” becomes useEffect—keeping only names, never source, and sending nothing off the computer. IraVoice has no GitHub Copilot integration: it types at the cursor and, by default, does not press Enter, so you still review and submit the prompt in Copilot CLI yourself.