Prompt dictation
It is not a voice keyboard. You speak when you need to steer the agent.
GitHub
Open-source alpha for LLM coding CLIs
Listen to your coding agent and step in by voice the moment it matters.
Voice Copilot runs beside Claude Code, Codex, and other CLI agents. A separate commentator LLM compresses agent events into short spoken updates, so you stay in control without staring at the terminal.
Voice Copilot is for the moments when the agent is doing useful work, but you still need situational awareness: what it understood, what it changed, what risk it noticed, and whether you should step in.
The agent found a queue timing issue. It will keep the public API stable and add compatibility coverage first.
The demo is published on YouTube, so visitors can understand the product before reading docs.
It is not a voice keyboard. You speak when you need to steer the agent.
It does not read every line aloud. It summarizes the important events.
The underlying CLI stays in control. Voice Copilot is the companion layer.
Use stream-json where CLIs expose events, HTTP proxy mode where they do not, and PTY fallback when needed.
Ask a side question, queue a correction for the next turn, or copy it for manual paste depending on CLI capability.
Choose cloud-light defaults or local TTS, STT, and commentator backends. Provider failures surface loudly.
Listening is the default, but the popup keeps a readable trace for deeper inspection.
Voice Copilot is currently a first public alpha. It is for advanced users who are comfortable testing CLI workflows and sharing practical feedback.
pipx install voice-copilot
voice-copilot serve
voice-copilot run claude -p "fix failing tests"
# Or with uv
uvx voice-copilot run codex -p "explain this repo"