Quick Start: Get your first Claw up and running in 30 minutes
💡 Goal: Install OpenClaw on your own computer and connect a model within 30 minutes, and receive the first reply from your first Claw in the browser Control UI. This is the first stop of the entire OpenClaw tutorial—beginners with zero experience can follow along all the way.
OpenClaw is an open-source AI assistant that runs on your own devices: a single Gateway process manages sessions, tools, and channel connections, while state, memory, and credentials all stay on your hardware; models are swappable plugins, and domestic models like DeepSeek, Qwen, Zhipu GLM, and Kimi are all first-class citizens. How do you use OpenClaw, and where should you start? Just follow this stage lesson by lesson.
🎯 Stage Goals
- Install and run OpenClaw:
openclaw gateway statusshows the Gateway is running (listening on port 18789 by default) - Connect at least one model: reuse an existing Claude Code / Codex CLI login, or configure an API Key from a model provider (DeepSeek, Qwen, GLM, Kimi, MiniMax, Doubao, Hunyuan, and Qianfan all have official plugins)
- Complete your first conversation:
openclaw dashboardopens the Control UI—send a message and receive an AI reply - Be able to rescue yourself if installation fails: first run
openclaw triagefor one-click diagnostics, then check the troubleshooting page against the error
⏳ Estimated Time
- Just want to get it running ASAP: Lessons 1.1 and 1.3 take about 18 minutes (1.2 covers what's new in 2.0 and can be caught up later). Of that, the hands-on part—installation + onboarding + first conversation—takes about 5 minutes by official estimates, provided you already have usable model credentials (Claude Code / Codex CLI login or API Key)
- Want to connect domestic models first: add the 1.6 model overview plus one provider-specific page (e.g., 1.7 DeepSeek), for a total of about 34 minutes—the extra ~16 minutes (1.6 + one provider page) buys you a complete configuration for one domestic model; API Key registration time not included
- Complete this stage in full (15 lessons, including what's new in 2.0, alternative installation, troubleshooting, and 9 model integration pages): about 132 minutes, or about 90–170 minutes with buffer
🚀 Before You Start
- A macOS / Linux / Windows computer. Windows has three paths: the native Windows Hub app, the PowerShell installer, or WSL2 Gateway
- Node.js 24.16.0+ (24.x) or 26.1.0+, Node 26 recommended. Don't have it? No problem: if Node is missing, the installer will install Node 26 on macOS and Node 24 LTS on Linux
- A "model pass": an existing Claude Code or Codex CLI login, or an API Key from any model provider—onboarding can detect and reuse it automatically
- Users in China: mainstream domestic models all have official provider plugins—DeepSeek, Qwen, Zhipu GLM, Kimi, MiniMax, Doubao/Volcano Engine, Hunyuan, Qianfan; 1.7–1.13 walks through each provider's configuration
- On a tight budget? 1.15 local models teaches you to run models on your own machine with Ollama or vLLM, with no API spending
- A cup of water ☕—from typing the install command to receiving your first reply, it's faster than you think
💡 Already installed OpenClaw?
Jump straight to 1.6 Model Connection Overview to switch models, or go to Daily Use to learn the Control UI and chat commands. Stuck on an error? Start with What to Do If Installation Fails.
📚 Lessons in This Stage
| Lesson | Content | Duration |
|---|---|---|
| 1.1 OpenClaw What is it | Open-source self-hosted AI assistant, Gateway architecture, what it can do, is it free, old names Clawdbot / Moltbot | 8 minutes |
| 1.2 OpenClaw What's new in 2.0 | What changed in v2026.8.1 (i.e., OpenClaw 2.0): brand-new Control UI, cross-session memory, Skills, native app | 10 minutes |
| 1.3 Installation: up and running in 5 minutes | One-command install, onboarding walkthrough, Gateway installed as a background service, first conversation in the Control UI | 10 minutes |
| ↳ 1.4 Alternative installation methods | Desktop apps (Windows Hub, macOS menu bar), Docker, direct npm install | 8 min |
| 1.5 What if the installation fails | Installation and startup troubleshooting: Node version, Gateway won't start, update failures | 8 min |
| 1.6 Connecting models overview | provider system, provider/model reference syntax, openclaw models command, switching models within a chat | 8 min |
| ↳ 1.7 Connecting DeepSeek | V4 series configuration; old IDs such as deepseek-chat were retired on 2026-07-24 and must be replaced with new ones | 8 min |
| ↳ 1.8 Connecting Tongyi Qwen | DashScope three plans (Coding / Standard / Token Plan), default qwen3.5-plus | 8 min |
| ↳ 1.9 Connecting Zhipu GLM | Z.AI dual endpoints, Coding Plan, glm-5.3 | 8 min |
| ↳ 1.10 Connecting Kimi | Distinguish between the two plugins: Moonshot and Kimi Coding, default kimi-k3 | 10 min |
| ↳ 1.11 Connecting MiniMax | M3 series, two ways to connect: Coding Plan (OAuth) and API Key | 8 min |
| ↳ 1.12 Connecting Doubao / Volcengine | Ark endpoint, Coding Plan default ark-code-latest | 8 min |
| ↳ 1.13 Connecting Hunyuan and Qianfan | Tencent TokenHub (hy4-preview), Baidu Qianfan (ernie-5.1) | 10 min |
| ↳ 1.14 More domestic models and relays | Models without official plugins (Spark, Baichuan, etc.), OpenRouter / LiteLLM, custom relay endpoints | 10 min |
| ↳ 1.15 Local models | Ollama, vLLM, ds4: run models on your own machine | 10 min |
🎒 What you will gain
- A properly installed OpenClaw Gateway that runs in the background, plus a Claw that can chat — all later stages build on top of it
- The architectural intuition that "the Gateway is the trust boundary, while models and agent harnesses are swappable plugins," so config files no longer look confusing
- Access configuration for at least one model: environment variables, endpoints, and
provider/modelreferences done right in one pass - A roadmap after getting it running: go to Daily Use to learn the memory system and chat commands, and go to Scenario Practice to hook Claw into WeCom and Feishu