Install and quick start
This page takes you from nothing to a working AI employee: pick an AI brain, install DuDuClaw, create your first employee in the dashboard, then connect a chat app. It is condensed from the DuDuClaw README.
1. Pick an AI brain
Section titled “1. Pick an AI brain”DuDuClaw does not ship a model of its own. It provides the plumbing: channels, memory, tasks and the cost ledger. Pick one of these (you can also do it later in the browser wizard):
- An AI command-line tool with an API key, such as Claude Code, Codex, Antigravity or Grok. The full list is in multi-runtime execution. Anthropic and Google do not allow third-party products to use the sign-in tokens of personal accounts, and accounts have been suspended for it, so use an API key.
- An API key for any OpenAI-compatible service.
- A local model behind an OpenAI-compatible server (llama-server, Ollama, vLLM, llamafile). No cloud account needed.
Gemini CLI is deprecated; new setups should use Antigravity. The schedule is in deprecations and removals.
2. Install
Section titled “2. Install”Desktop app (recommended for personal use)
Section titled “Desktop app (recommended for personal use)”The desktop app starts the local gateway when you open it, so you never need a terminal. It shares ~/.duduclaw with the command-line version. Download the newest desktop-v… release from GitHub Releases:
| Platform | File | Notes |
|---|---|---|
| macOS (Apple Silicon / Intel) | DuDuClaw_*.dmg |
Signed and notarized by Apple |
| Windows x64 | DuDuClaw_*_x64_en-US.msi |
No Authenticode certificate yet, so SmartScreen shows an “unknown publisher” warning once; choose “More info”, then “Run anyway” |
| Linux | *_amd64.AppImage / .deb |
No signing needed |
npm (servers, or if you prefer the command line)
Section titled “npm (servers, or if you prefer the command line)”You only need Node.js 20 or later:
npm install -g duduclawnpm installs a prebuilt binary for your platform (macOS ARM64/x64, Linux x64/ARM64, Windows x64). No compiler, no Rust. If the install asks you for Rust or MSVC Build Tools, you are on the build-from-source path meant for contributors; use the line above instead.
Build from source
Section titled “Build from source”Needs Rust 1.85+ and Node.js 20+:
git clone https://github.com/zhixuli0406/DuDuClaw.gitcd DuDuClawcd web && npm ci --legacy-peer-deps && npm run build && cd ..cargo build --release -p duduclaw-cli -p duduclaw-gateway --features duduclaw-gateway/dashboard./target/release/duduclaw run3. Start it and create your first AI employee
Section titled “3. Start it and create your first AI employee”With the desktop app, just open it. With npm or a source build:
duduclaw run # starts the gateway, channels, scheduler and dispatcherThen open http://localhost:18789 in a browser. The first visit shows a setup wizard: pick an AI backend, create your first AI employee, and chat with it in the built-in WebChat. You do not need to run duduclaw onboard in a terminal first. Where every setting lives is mapped in the dashboard settings map.
4. Connect a chat app
Section titled “4. Connect a chat app”Paste a bot token on the dashboard’s Channels page and the same AI employee appears on Telegram, LINE, Discord and the rest, without a restart. DuDuClaw supports 11 channels: Telegram, LINE, Discord, Slack, WhatsApp, Feishu, Google Chat, Microsoft Teams, WeCom, DingTalk and WebChat.
- Webhook channels such as LINE and WhatsApp need a public HTTPS URL. Tailscale Funnel, ngrok and Cloudflare Tunnel setups are in the deployment guide.
- Registering Google Chat and Microsoft Teams bots is covered in Google Chat and Teams channels.
5. Hand over the first job
Section titled “5. Hand over the first job”In any connected channel, type:
/goal Summarise this week's complaints || List the top three issues with a count for eachThe part after || is the acceptance criteria. When the AI employee reports it is done, a judge reviews the result for correctness, completeness and safety (simple one-step goals get correctness and safety only); it counts as done only when every check passes, and a stuck task comes back to you. Details are in the autonomous goal loop and the goal loop guide.
Useful next steps
Section titled “Useful next steps”duduclaw agent create # create more AI employeesduduclaw wizard # set up from an industry templateduduclaw status # system health snapshotduduclaw update # check for and install updatesduduclaw service install # start on boot (launchd / systemd)Moving over from OpenClaw, Hermes or paperclip? See migrating from other platforms.