Lumi 是 AI Skill Hub 本期精选Agent工作流之一。综合评分 7.5 分,整体质量较高。我们推荐使用将其纳入你的 AI 工具库,帮助提升工作效率。
A voice-first AI Agent living in your macOS menu bar。语音AI助手,住在你菜单栏。
Lumi 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
A voice-first AI Agent living in your macOS menu bar。语音AI助手,住在你菜单栏。
Lumi 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
# 方式一:npm 全局安装 npm install -g lumi # 方式二:npx 直接运行(无需安装) npx lumi --help # 方式三:项目依赖安装 npm install lumi # 方式四:从源码运行 git clone https://github.com/Wechat-ggGitHub/Lumi cd Lumi npm install npm start
# 命令行使用
lumi --help
# 基本用法
lumi [options] <input>
# Node.js 代码中使用
const lumi = require('lumi');
const result = await lumi.run(options);
console.log(result);
# lumi 配置说明 # 查看配置选项 lumi --config-example > config.yml # 常见配置项 # output_dir: ./output # log_level: info # workers: 4 # 环境变量(覆盖配置文件) export LUMI_CONFIG="/path/to/config.yml"
<p align="center"><strong>The AI Agent that lives in your menu bar.</strong></p>
<p align="center">Voice-first. Always ready. Knows you, remembers you.</p>
<p align="center"> <img src="https://img.shields.io/badge/macOS-13%2B-informational" /> <img src="https://img.shields.io/badge/Electron-35-blue" /> <img src="https://img.shields.io/badge/Next.js-15-black" /> <img src="https://img.shields.io/badge/license-MIT-green" /> </p>
<p align="center"> <img src="assets/banner.png" alt="Lumi Banner" width="800" /> </p>
<p align="center"><strong>English</strong> | <a href="./README_CN.md">中文</a></p>
Lumi is a native macOS voice AI assistant. It lives in your menu bar — you speak, it listens, it gets things done. Search the web, write code, control your system, and it remembers everything you've talked about.
We believe the most natural way to interact with AI is voice — you speak, it listens. And the best way for AI to respond is through vision — it shows you. Karpathy said that voice is the natural input for humans, and vision is the natural output for machines — nearly a third of the brain's processing power is dedicated to visual information. Lumi is still an early demo, but we want to keep refining in this direction, exploring what the interaction between humans and truly intelligent AI should look like.
The Electron main process spawns a Next.js 15 standalone server on a random port, connecting front-end and back-end via IPC (not REST API). In production, Next.js runs as a child process inside Electron.
┌─────────────────────────────────────────┐
│ Electron Main │
│ │
│ ┌─────────────┐ ┌──────────────────┐ │
│ │ Tray + │ │ Voice Pipeline │ │
│ │ Shortcuts │ │ (ASR/TTS/VAD) │ │
│ └─────────────┘ └──────────────────┘ │
│ │
│ ┌─────────────────────────────────────┐│
│ │ Next.js 15 (embedded) ││
│ │ BrowserWindow ←→ IPC ←→ Main ││
│ └─────────────────────────────────────┘│
└─────────────────────────────────────────┘
git clone https://github.com/Wechat-ggGitHub/Lumi.git
cd Lumi
npm install
npm run electron:dev
npm run electron:build
The build output is in the release/ directory.
Some clips in the videos are sped up for demonstration. Actual Agent execution takes time.
https://github.com/user-attachments/assets/fc44c44b-82ee-4923-b701-030fe7c096b4
https://github.com/user-attachments/assets/de213cfb-6a36-4bae-a5bc-e74c91fc3015
AudioListener → WakeWordEngine (sherpa-onnx)
→ VoiceEndpoint (VAD silence detection)
→ AudioRecorder (recording)
→ ASR Provider (Volcengine / Alibaba Bailian)
→ Claude Agent SDK (AI processing)
→ TTS Provider (Volcengine / Alibaba Bailian)
→ SubtitlePopup (subtitle overlay)
Lumi是一个开源的AI工作流,提供了语音AI助手功能,虽然star数较少,但仍然值得关注。
该工具未明确声明开源协议,商业使用前请联系原作者确认授权范围,避免侵权风险。
AI Skill Hub 为第三方内容聚合平台,本页面信息基于公开数据整理,不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后,再部署至生产环境,并做好必要的安全评估。
经综合评估,Lumi 在Agent工作流赛道中表现稳健,质量良好。如果你已有明确的使用需求,可以直接上手体验;如果还在评估阶段,建议对比同类工具后再做决策。
| 原始名称 | Lumi |
| Topics | workflowtypescript |
| GitHub | https://github.com/Wechat-ggGitHub/Lumi |
| 语言 | TypeScript |
收录时间:2026-05-23 · 更新时间:2026-05-23 · License:未公布 · AI Skill Hub 不对第三方内容的准确性作法律背书。
选择 Agent 类型,复制安装指令后粘贴到对应客户端