与你的Rabbit R1中的任何AI模型对话。 在R1上使用GPT-5.2、Claude、Gemini、Qwen3.5或其他任何模型作为实时语音助手，支持LiveKit。默认情况下，语音使用GPT-4.1-mini / Cartesia。

搭配local-voice-ai，在自己的硬件上本地运行所有内容，无需云服务商。

这是什么

SKILL.md 是一个结构化提示，任何 AI 编码助手都可以从零开始构建完整的 R1 语音代理项目。它支持 Claude 代码、光标、Windsurf、Copilot、Aider 或任何基于大型语言模型的编码工具——只需输入提示词，让它自行构建即可。

最终项目包括：

• 一个Python语音代理（通过lk app create支架）运行STT -> LLM -> TTS
• 一个web服务器（web-server.py），服务R1前端并提供LiveKit令牌端点
• R1“创建”（creation/index.html）——运行在设备上的240x282像素语音界面
• 一个安装页面（creation/install.html），带有用于侧载到R1的二维码

如何使用

安装技能

这会安装SKILL.md，这样你的AI编码助手就能自动使用。

Rabbit R1 LiveKit Voice Agent

Talk to any AI model from your Rabbit R1. Use GPT-5.2, Claude, Gemini, Qwen3.5, or any other model as a real-time voice assistant on the R1 powered by LiveKit. By default, uses GPT-4.1-mini / Cartesia for voice.

Pair it with local-voice-ai to run everything locally on your own hardware w/ no cloud providers needed.

What This Is

SKILL.md is a structured prompt that any AI coding assistant can follow to build a full R1 voice agent project from scratch. It works with Claude Code, Cursor, Windsurf, Copilot, Aider, or any LLM-powered coding tool – just feed it the prompt and let it build.

The resulting project includes:

• A Python voice agent (scaffolded via lk app create) that runs STT -> LLM -> TTS
• A web server (web-server.py) that serves the R1 frontend and provides a LiveKit token endpoint
• An R1 “creation” (creation/index.html) – the 240x282px voice UI that runs on the device
• An install page (creation/install.html) with a QR code for sideloading onto the R1

How to Use

Install the Skill

npx skills add ShayneP/rabbit-r1-livekit-skill

This installs SKILL.md so your AI coding assistant can use it automatically. Works with Claude Code, Cursor, Windsurf, Copilot, Aider, and other tools that support the skills format.

Once installed, just tell your assistant what you want:

“Build me a voice assistant for my Rabbit R1 that answers trivia questions.”

In Claude Code, you can also invoke it directly:

/rabbit-r1-livekit

Manual Usage (Without Installing)

If you prefer not to install, copy the contents of SKILL.md into your assistant’s context (paste it into chat, add it as a reference file, etc.) and describe what you want.

What the Prompt Guides Your Assistant Through

1. Install the LiveKit CLI (lk) if needed
2. Scaffold the agent with lk app create --template agent-starter-python agent
3. Set up LiveKit Cloud credentials via lk cloud auth
4. Create the web server and R1 frontend files
5. Start all services and open an HTTPS tunnel
6. Provide an install link for the R1

Architecture

R1 WebView
  --> web-server.py (port 8081, HTTPS via tunnelmole)
    --> LiveKit Cloud (WebRTC SFU + Agent Dispatch)
      --> Python Agent (STT -> LLM -> TTS)

Flow:

1. User taps the R1 screen (required for mic access)
2. Frontend fetches a LiveKit token from /getToken
3. Token includes a RoomConfiguration that tells LiveKit which agent to dispatch
4. Frontend connects to a LiveKit room via WebRTC and publishes mic audio
5. LiveKit Cloud dispatches the named agent into the room
6. Agent processes audio through STT -> LLM -> TTS and publishes audio back
7. R1 plays the response through its speaker

Project Structure (After Running)

project-root/
  web-server.py              # Static file server + /getToken token API
  creation/