# openclaw-docs: a 767 page Chinese documentation site for the OpenClaw agent platform

> openclaw-docs is a VitePress site that documents the self hosted OpenClaw AI assistant platform for Chinese readers, split into an installation track, an engineering track, an AI framework track and a channel adapter track. It is a documentation repository rather than a program, and it is published as a static site on Azure.

**yeuxuan/openclaw-docs** — OpenClaw 中文文档站 | AI 智能体框架 源码剖析 安装教程 | WhatsApp Telegram Discord 飞书   多通道机器人

- Repository: https://github.com/yeuxuan/openclaw-docs
- Website: https://openclaw-docs.dx3n.cn
- Stars: 750 · Forks: 108
- Language: JavaScript
- License: not declared
- Published: 2026-10-21 · Updated: 2026-10-21 · Language: en
- Canonical page: https://hysenlabs.com/projects/yeuxuan-openclaw-docs

## A documentation repository, not a runtime

openclaw-docs is a tutorial site for OpenClaw, a self hosted multi channel AI assistant platform that connects a web control interface, chat channels, nodes, tools and models through a Gateway. The project describes its own purpose plainly: it is a tutorial site for Chinese readers whose goal is to explain the official architecture clearly enough that a beginner can get it running and then that a developer can read the source.

Everything in the repository serves that goal. The top level holds docs/, scripts/, package.json, package-lock.json, an AGENTS.md and a CLAUDE.md for coding agents, and a file named UPSTREAM_SYNC_2026-09-14.md which records a sync with upstream on that date. The docs/ directory holds the markdown, and the whole site is built by VitePress.

The manifest is minimal, with the package marked private and of type module, and only three dev dependencies, vitepress itself, the mermaid diagram library and a VitePress plugin for rendering mermaid. There are six scripts, four of which are the dev, build, preview and two audit commands described later. No runtime code, no test suite, no framework of its own. This is a site about a platform, and reading it tells you as much about the platform as it does about the documentation.

## The four tracks are the navigation, and the counts show the balance

The site claims 767 Chinese tutorials, divided into four learning tracks, and the distribution is the interesting part. Track 0 is installation and setup, covering a zero basis path, quick install, onboarding, the web control UI, nodes, AI provider access, channel access, Gateway operations and troubleshooting. Track 0 alone accounts for 680 pages.

Track A is the complete engineering line, 62 pages, covering the CLI startup framework, the Gateway control plane, plugins and channel adapters, nodes, routing and session keys, the agent execution chain and function level source analysis. Track B is the AI core framework, 24 pages, on context engineering, the agent state machine, tool policy and approval, model fallback, the memory system and the hook plugin injection mechanism. Track C is the channel adapters, which cover the interface contract, the registration chain, account lifecycle, inbound routing and outbound send decoupling, and which is counted inside Track A rather than separately.

A ratio of 680 installation pages against 62 engineering pages says something about who this is for. The intended reader is someone who has not run the platform yet and needs a path, and the site is built as a guided course rather than a reference manual, with each track linking to its own section of the built site.

## The docs/ tree shows the shape of the platform as much as the docs do

The directory listing in the README is effectively a map of OpenClaw itself, and it is worth reading as such rather than as a list of folders. Under tutorials/ there is getting-started, installation for Docker, Node and cloud server deployment, gateway, channels, providers, concepts, tools, plugins, cli, reference, platforms, diagnostics, plan, web, nodes, automation and help.

The plan/ directory is described as holding architecture plans and maintainer design drafts, which is a different kind of document from a tutorial and suggests the site follows upstream work in progress rather than only shipping stable versions. The diagnostics/ and help/ directories split troubleshooting from frequently asked questions, and the reference/ directory sits next to the topical chapters, so the site mixes a course structure with a reference structure.

Two entries in the tree are outside the documentation itself. scripts/convert-mdx.mjs is a batch converter from MDX to VitePress Markdown, which implies the content was authored somewhere else and brought across, and scripts/ping-indexnow.mjs pushes URLs to Bing IndexNow after a build, which is why the site can say it has automatic submission. The site configuration lives in docs/.vitepress/config.mts for SEO, navigation and the sidebar, with a custom theme alongside it.

## Running the site locally is four npm commands

The quick start is short, and the tooling matches it. You clone the repository, move into the directory, install the dependencies and start the VitePress dev server.

```bash
git clone https://github.com/yeuxuan/openclaw-docs.git
cd openclaw-docs
npm install
npm run docs:dev
```

The dev script runs vitepress dev docs with the host set to 0.0.0.0, so the server is reachable from other machines on the network rather than only from the loopback interface, which is a small convenience when you are checking the site on a phone or a second machine. Opening the local address printed in the terminal gives you the whole site.

For a production build there are two more commands, npm run docs:build and npm run docs:preview. The build script is vitepress build docs, and the preview script serves the built output so you can check the result before publishing. The published site runs on Azure Static Web Apps, which is the reason the build output is nothing but a directory of static files with no server component behind it.

## Two audit scripts suggest the site is checked by machine as well as by hand

The scripts list contains two commands that have nothing to do with rendering. docs:audit-humanizer runs node scripts/audit-humanizer.mjs and docs:audit-seo runs node scripts/audit-seo.mjs. Neither of those two files appears in the scripts/ listing in the README, which shows only the MDX converter and the IndexNow pinger, so the audits are either newer than the documented tree or live elsewhere.

Their names suggest what they check. An audit for a humanizer looks for patterns of machine written prose, and an audit for SEO looks at the page level metadata, which fits a site that claims 767 pages with individual descriptions, a sitemap and automatic IndexNow submission. A documentation site that large cannot be checked by reading, so the approach is to have a script flag the pages that look wrong.

This is a genuinely useful idea to copy even if you never read the tutorials, because the failure mode of a large generated documentation set is uniform, unedited prose that reads the same on every page. Whether the audits are strict or advisory is not documented, and neither is what happens when one fails, so for a contributor the answer is to run them before opening a pull request rather than to find out in review.

## What the documentation says about the platform it documents

Since the subject is only described through the tutorial site, the platform description is worth setting out. The subject is an open source personal AI assistant that lives on hardware you control and does its work through a browser, a chat application and mobile nodes. One component, the Gateway, unifies the web control interface, the messaging channels, the nodes, the tools and the models.

The supported chat platforms listed are WhatsApp, Telegram, Discord, Slack, Signal, iMessage and BlueBubbles, Feishu, Mattermost, Google Chat, Microsoft Teams, Matrix, LINE, Zalo, QQ, WeChat, IRC, Nostr, Twitch and WebChat. The supported model providers listed are Anthropic Claude, OpenAI GPT, DeepSeek, Qwen, Kimi, Zhipu GLM, MiniMax, Ollama for local models, OpenRouter, LiteLLM, Cloudflare AI Gateway, Vercel AI Gateway, NVIDIA, Together AI, HuggingFace, vLLM and SGLang.

That breadth is the point of the Gateway design as the site presents it, and it is also the reason the adapter track exists, since each channel needs its own registration, account lifecycle and message routing. The official project is at github.com/openclaw with English documentation at docs.openclaw.ai, and this site is explicitly the Chinese counterpart rather than a replacement.

## The licence is the loose end, and it is a documentation project

The README carries an MIT licence badge, while the project is listed with an unknown licence. The two can be reconciled, since a badge can point at a licence that is in the repository, but the top level listing does not include a LICENSE file at all. The entries are .github/, .gitignore, AGENTS.md, CLAUDE.md, README.md, the upstream sync file, docs/, package.json, package-lock.json and scripts/.

For a documentation project that is the one thing worth raising, because documentation is the content. If you want to translate these tutorials, quote them at length or mirror them, and the file that would answer the question is not visible in the tree. The safe route is to ask the author, and if you are only reading the site, the question does not arise.

There is also a practical point about the upstream relationship. The site tracks a platform it does not own, and the UPSTREAM_SYNC file with a date in its name shows that syncing is a deliberate, periodic task rather than an automatic one. Between syncs, a page can describe a version of OpenClaw that is not the version on your server, so the sync file is the closest thing the repository has to a currency marker.

## Conclusion

Adopt openclaw-docs if you read Chinese and want a mapped route through OpenClaw before touching the code, since the track structure tells you which page answers which question and the install track is far more granular than the official documentation. Do not adopt it as a substitute for the upstream documentation, because it is a third party site that mirrors a project maintained elsewhere, and the version it describes may not be the version you install. Before you rely on it, check the UPSTREAM_SYNC file at the repository root for how recently it was synced with upstream, and remember that the badge claims an MIT licence while the root listing shows no LICENSE file, so confirm the terms with the author before redistributing the text.

## FAQ

### What is openclaw-docs and who is it for?

It is a VitePress documentation site with 767 Chinese tutorials on the OpenClaw self hosted AI assistant platform, built for readers who want to get the platform running and then understand its source. The content is split into an installation track of 680 pages, an engineering track of 62 pages, an AI framework track of 24 pages, and a channel adapter track counted inside the engineering track.

### How do I run the openclaw-docs site locally?

Clone the repository, change into the directory, run npm install and then npm run docs:dev, which starts VitePress on the docs folder with the host set to 0.0.0.0. The local address printed in the terminal serves the full site.

### How is the openclaw-docs site built and deployed?

npm run docs:build runs vitepress build docs and writes the static output to docs/.vitepress/dist/, and npm run docs:preview serves that output locally. The site is deployed on Azure Static Web Apps, and a script in scripts/ pushes URLs to Bing IndexNow after a build.

### What does the openclaw-docs site say about OpenClaw itself?

It presents OpenClaw as software you host yourself and reach from a browser, a chat application or a mobile node, with a Gateway tying the control interface, the messaging channels, the nodes and the model providers together. The channels named include WhatsApp, Telegram, Discord, Slack, Signal, Feishu, Matrix, QQ, WeChat and IRC, and the provider list runs from Anthropic Claude and OpenAI GPT through Ollama and LiteLLM to several gateway services.

### Can I reuse or translate the openclaw-docs content?

The README shows an MIT licence badge, but the project is listed with an unknown licence and no LICENSE file appears among the top level entries, which are .github, .gitignore, AGENTS.md, CLAUDE.md, README.md, an upstream sync file, docs, package.json, package-lock.json and scripts. Ask the author for the exact terms before redistributing or translating the text.

## Sources

- [Issues](https://github.com/yeuxuan/openclaw-docs/issues)
- [Project website](https://openclaw-docs.dx3n.cn)
- [README](https://github.com/yeuxuan/openclaw-docs/blob/main/README.md)
- [yeuxuan/openclaw-docs on GitHub](https://github.com/yeuxuan/openclaw-docs)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/yeuxuan-openclaw-docs