OpenClaw Voice Control

OpenClaw Voice Control is a local macOS voice-control integration for OpenClaw.

Repository source:

• GitHub: CarrotYuan/openclaw-voice-control

It provides:

• wakeword activation
• local microphone capture
• local ASR with FunASR / SenseVoice
• forwarding recognized text to a local OpenClaw agent
• macOS TTS playback
• optional overlay UI
• launchd-based background resident behavior
• auto-start after user login

Platform

• macOS only

Safety Warning

This skill adds a voice entrypoint, and that entrypoint is not identity-bound.

That means:

• anyone near the microphone may be able to trigger it
• any capability exposed through the connected OpenClaw agent may become reachable by voice

Recommended safety practice:

• add explicit safety constraints to the connected agent prompt
• require confirmation for high-risk actions
• avoid broad autonomous permissions for the voice-facing agent
• use least privilege for tools and connected systems

Main Path

Treat this as a local deployment skill, not as a prompt-only helper.

When this skill is installed into OpenClaw:

• work inside the current installed skill directory for the current conversation agent
• use the declared GitHub repository as the source of truth
• do not silently switch to another local clone or an already-prepared environment
• do not continue until the workspace contains the real repository files, not just SKILL.md

The main installation guide is the repository README.md. README.zh-CN.md is the Chinese companion guide.

Use this as the standard install path:

1. sync the full repository into the current installed skill workspace
2. create and activate .venv
3. run pip install -e .
4. download or populate the SenseVoice model directory
5. download or populate the VAD model directory
6. copy .env.example to .env
7. fill the required values in .env
8. use the default openWakeWord route
9. run direct-run validation with both the voice service and overlay process
10. ask whether the user wants background resident behavior and auto-start
11. if yes, run ./scripts/deploy_macos.sh
12. if no, stop after direct-run validation

Before running any system-changing step in that path, explicitly tell the user what you are about to do and get confirmation for:

• fetching and checking out the repository into the skill workspace
• running pip install -e . from the repository
• downloading large local models or wakeword assets
• enabling background resident behavior or launchd auto-start

Minimum command path: