All the power of Claude Code, no terminal required. Write code, control browsers, create Digital Humans — your AI, on standby around the clock.
Halo is built on top of Claude Code, with a complete product layer totaling over 300,000 lines of code, validated by tens of thousands of users, and running stably in enterprise environments. On top of that, Halo also delivers:
| What the terminal can't do | Halo can |
|---|---|
| See every file AI generates | Artifact Rail previews code, HTML, and images in real time |
| Stops when you leave the computer | Remote Access — continue anytime from phone / H5 / WeChat / Android client |
| Have to start manually every time | Digital Humans run automatically 7x24 |
| Let non-technical colleagues use it | Download and go, zero configuration |
| Automate browser operations | AI Browser — embedded browser directly controlled by AI |
Powered by Claude Code — 100% compatible with Claude Code's Agent capabilities, MCP, and Skills.
Most AI tools require you to sit in front of the screen, going back and forth in conversation. Halo is different — it can work on its own, and you only need to make decisions at key checkpoints.
Create a Digital Human, give it a task and an execution frequency, and it will run autonomously on schedule:
- Push a tech news digest every morning
- Check online service status every hour and notify you of anomalies
- Run competitive analysis on a schedule and generate comparison reports
- Monitor GitHub dependency updates and security vulnerabilities
- Track keyword mentions across social media
Install with one click from the Digital Human Store, or create your own using natural language.
Think of it as a cron job + AI Agent hybrid — except you just speak in plain language.
Digital Humans have the exact same Agent capabilities as conversation mode — the same Claude engine, MCP toolchain, and AI Browser — they just trigger automatically on schedule without needing you at the computer.
WeChat is your control panel. Digital Humans support two-way conversational control via personal WeChat / WeCom (Enterprise WeChat) — not just receiving notifications, you can give instructions to Digital Humans, check progress, and request reports directly in WeChat.
Typical AI browser automation has the AI fumble around figuring out what to click and fill every time, which frequently fails.
Browser Skill takes a different approach: pre-write reusable scripts for common operations on each website. The AI only needs to decide "which script to call now" — the script already handles the specifics of how to operate the website.
Skill scripts run directly in a real browser environment via Halo's browser_run — with access to the page DOM, cookies, and internal APIs, just like operating in the Chrome DevTools console. For example, here is the core code of a Bilibili notification reading Skill:
// .claude/skills/bili-get-messages/index.js
async (params) => {
// Directly call Bilibili's internal API — cookies are automatically included, no extra authentication needed
const resp = await fetch('https://api.bilibili.com/x/msgfeed/reply?platform=web', {
credentials: 'include'
}).then(r => r.json())
// Return structured data to the AI, which decides how to respond
return {
success: true,
notifications: resp.data.items.map(item => ({
user: item.user.nickname,
comment: item.item.source_content,
video_title: item.item.title
}))
}
}When a Digital Human calls it, all it takes is: browser_run({ file: "skills/bili-get-messages/index.js" }) — after getting the data, the AI decides which items need replies and how to reply.
For example, a Zhihu Digital Human's workflow:
- AI decides: time to check for new invited questions
- Calls
zhihu-creator-invitedSkill -> script automatically fetches the invitation list and returns structured data - AI judges: this question is worth answering, starts writing
- Calls
zhihu-publish-answerSkill -> script automatically fills the editor and publishes
AI makes the decisions, Skills handle the operations. Stable, repeatable, reliable.
There are already ready-made Skills for platforms like Bilibili, Zhihu, WeChat, Xiaohongshu, and more. The community can also contribute their own.
Once Remote Access is enabled, your phone / H5 / WeChat / Android client can all control the Halo on your desktop. During meetings, commuting, or even from a hospital bed (true story), check AI's work progress anytime and issue new instructions.
Get started in 30 seconds:
- Download and install, launch Halo
- Enter your API Key (Anthropic recommended)
- Start chatting — try
Build a todo app with ReactorHelp me analyze the code structure of this project - Watch files appear in the Artifact Rail, click to preview, request changes
Recommended models: Claude Sonnet / Opus series
| Platform | Download | Requirements |
|---|---|---|
| macOS (Apple Silicon) | .dmg | macOS 11+ |
| macOS (Intel) | .dmg | macOS 11+ |
| Windows | .exe | Windows 10+ |
| Linux | .AppImage | Ubuntu 20.04+ |
| Android | .apk | Android 8+ |
| iOS | Build from source | iOS 15+ |
Download, install, run. No Node.js, no npm, no terminal needed.
git clone https://github.com/openkursar/hello-halo.git
cd hello-halo
npm install
npm run prepare
npm run dev|
Open the Digital Human Store, pick one, fill in a few configuration fields, and it starts running automatically. No coding required, no prompts to write.
|
Write a You can also write Browser Skills ( |
Remote Access: Control Halo from anywhere
AI Browser
ai_brower_480.mp4
┌──────────────────────────────────────────────────┐
│ Halo Desktop │
│ │
│ React UI <─IPC─> Main Process <──> Claude │
│ (Renderer) ┌───────────┐ Code SDK │
│ │ Digital │ (Agent │
│ │ Humans │ Loop) │
│ │ Scheduler │ │
│ └───────────┘ │
│ │ │
│ ~/.halo/ (local) │
└──────────────────────────────────────────────────┘
- 100% Local — Your data never leaves your machine (except API calls)
- No Backend Required — Pure desktop client, use your own API Key
- Agent Loop — Tool execution, not just text generation
- Space System — Isolated workspaces, projects don't interfere with each other
- Skills — Install skill packs to extend Agent capabilities
- AI Browser — Embedded CDP browser, AI directly controls web pages
- Multi-Model Support — Anthropic, OpenAI, DeepSeek, and any OpenAI-compatible API
- Dark/Light Themes — Follows system preference
- Multi-Language — Chinese, English, Spanish, and more
- Claude Code SDK Agent Loop
- Space and Conversation Management
- Artifact Preview (Code, HTML, Images, Markdown)
- Remote Access
- AI Browser (CDP)
- MCP Server Support
- Skills System
- Digital Humans and Digital Human Store
- Third-party Ecosystem Plugin Compatibility
- Enhanced Code Editing Experience
- Visual Git + AI-Assisted Code Review
- AI-Powered File Search
git clone https://github.com/openkursar/hello-halo.git
cd hello-halo
npm install
npm run prepare
npm run dev- Translations —
src/renderer/i18n/ - Bug Reports — Issues
- Feature Suggestions — Discussions
- Code Contributions — PRs welcome
See CONTRIBUTING.md for details.
If the QR code has expired, add WeChat: go2halo with the note "Halo"
In October 2025, a simple frustration: I wanted to use Claude Code, but I was stuck in meetings all day.
During a boring meeting, I thought: What if I could control Claude Code on my home computer from my phone?
Then came the second problem — non-technical colleagues wanted to use it too, but got stuck at installation. "What's npm?"
So I built Halo: a visual interface, one-click install, remote access. The first version took a few hours. Everything after that? 100% built by Halo itself.
Now, we believe the next step is the AI Workstation: AI no longer needs someone watching to get work done. You set the goals, Digital Humans push forward autonomously 7x24. Writing code, running tests, monitoring deployments, generating reports — running continuously, with you only making decisions at key checkpoints.
That's what Halo is building.
MIT — LICENSE