Open-source project
jwangkun/hermes-agent-guide avatar
jwangkun/hermes-agent-guide

hermes-agent-guide: A 16-Volume Chinese Whitepaper on the Hermes Agent Framework

Hermes Agent 是目前开源社区最具潜力的 AI Agent 框架之一。它继承了 OpenClaw 的优秀基因,在架构设计、记忆系统、技能生态和自动化能力上实现全面升级。但 Hermes 的文档分散在多个仓库和平台,缺乏一本系统性的中文指南。 本书填补的正是这个空白。全书 16 册、30 万+ 字,覆盖从安装部署到高阶开发、从个人玩赚到企业服务、从单 Agent 操控到多 Agent 编排的完整知识图谱。

702 stars150 forksPythonNOASSERTION

At a glance

What is it?
The repository is a Markdown book, not a library. It documents Hermes Agent across 16 volumes, ships a Python script that converts the chapters to PDF, and splits its licence between CC BY-NC-SA 4.0 for prose and MIT for sample code.
Who is it for?
Adopt this guide if you read Chinese and want a structured path into Hermes Agent, particularly the four reading routes for beginners, developers, OpenClaw migrants and people evaluating commercial use. Do not adopt it if you need an English reference, or if you want to redistribute the prose commercially: the CC BY-NC-SA 4.0 terms on the book content rule that out, while the MIT terms cover only the sample code.
Can I use it commercially?
Check first. The repository uses a licence we do not classify automatically, so read its LICENSE file before any commercial use.
Is it still maintained?
Yes. The repository last received commits 2 days ago.
What is it written in?
Mainly Python, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.

DEEP OPEN-SOURCE ANALYSIS

What the Repository Actually Contains

This is not a software package. There is nothing to import. The repository is a book written in Markdown, with a Python script attached for producing a PDF. The README describes it as 16 volumes and more than 300,000 Chinese characters, and the table of contents lists files from 00-目录与导读.md through 16-未来展望与总结.md. The primary language field on the repository is Python, which reflects generate_pdf.py rather than any agent runtime. Anyone arriving expecting a Hermes Agent distribution will be looking in the wrong place.

The stated problem is documentation fragmentation. The README says Hermes documentation is spread across multiple repositories and platforms and that no systematic Chinese guide existed, so the book was written to fill that gap. The intended readers are listed explicitly: beginners who have never automated anything, developers who want the internal architecture, founders looking for commercial angles, analysts comparing frameworks, and enterprise decision makers assessing integration and total cost of ownership. That is a wide net, and the four reading routes exist to compensate: a 3 to 4 hour quick start, a 15 to 20 hour technical route, an 8 to 10 hour monetisation route, and a 6 to 8 hour migration route for existing OpenClaw users.

How the Reading Routes Sequence the Chapters

The route structure is the most concrete design decision in the repository, and it is worth reading as a dependency graph rather than a marketing device. The quick start route runs 01, 03, 05, 06 and a selection from 12, which means it deliberately skips the architecture volume (04) and the memory volume (07). The technical route inserts 02, 04, 07, 08, 09, 11 and 13. The migration route for OpenClaw users is built around the same spine but swaps in 13-OpenClaw对比与迁移.md as its marked centrepiece.

The chapter titles tell you what the author considers the load-bearing parts of Hermes: a five-layer architecture, a three-layer memory system (session memory, persistent memory, and skill-level memory), a skill lifecycle with automatic creation and a Skills Hub, 47 built-in tools grouped into seven categories, and MCP support described as connecting to more than 6000 MCP services alongside cron scheduling and multi-agent orchestration. Volume 10 covers integration with 15 or more platforms including Feishu, WeChat, Discord and Slack. Volume 14 breaks out nine monetisation paths. These are claims about Hermes made by the guide, not claims this review can verify, since the guide itself is the only material available here.

Building the PDF from the Markdown Sources

The only executable artefact documented in the README is the PDF pipeline. Two dependencies are named, markdown and weasyprint, and the invocation is two lines:

pip install markdown weasyprint python generate_pdf.py

The script is described as converting the Markdown chapters into a PDF named Hermes-Agent白皮书-养马从入门到精通-鲲鹏Talk.pdf. Nothing in the supplied material describes the script's internals, whether it accepts arguments, how it locates the chapter files, or what happens if a chapter is missing. If you intend to run it, read generate_pdf.py before executing it rather than assuming the two-command example is the whole story.

For the Hermes Agent itself, the guide points at volume 05 for installation and volume 06 for first run and model configuration, and the README does not reproduce those commands. That is a deliberate boundary: the repository documents Hermes rather than installing it. If you want the actual install commands, they live inside 05-安装部署全攻略.md, which the README describes as covering three approaches (local, Docker and VPS). This review cannot confirm what those commands are, because the chapter bodies are not part of the supplied material.

The Split Licence Is the Constraint That Matters Most

The repository's licence field reads NOASSERTION, and the README explains why: two different licences apply to two different things. The book content is under CC BY-NC-SA 4.0, which permits sharing and adaptation with attribution, but not commercial use, and requires derivatives to carry the same licence. The sample code is under MIT, which permits commercial and non-commercial use.

The practical consequence is that the prose cannot be folded into a paid course, a commercial internal handbook, or a product's documentation without permission, even though the code snippets inside it can be reused freely. That is an unusual split for a technical book and it is easy to get wrong, because a reader who copies a chapter wholesale into a company wiki is copying prose, not code. The README states the split plainly, so the ambiguity is not hidden, but anyone repurposing the material needs to decide which licence applies to the specific extract they are using. This is a description of the stated terms, not legal advice.

There is no release history in the repository, so there is no changelog to consult when the underlying Hermes project changes. The most recent push recorded is 2026-04-22, which tells you when the repository was last touched but not which Hermes version the chapters describe.

Where the Guide Is Likely to Age Poorly

The guide's own framing creates its main risk. It describes Hermes as inheriting from OpenClaw and upgrading architecture, memory, skills and automation, and it devotes an entire volume to a 12-dimension comparison and migration path. That kind of content is tightly coupled to a moving target. A chapter enumerating 47 built-in tools, or listing 6000+ MCP services, or describing a three-layer memory system, is accurate for the snapshot it was written against and drifts as the framework changes. With no releases and no version pinning visible in the README, a reader has no signal about which snapshot each chapter reflects.

The second limitation is linguistic. Every chapter filename is Chinese, the reading routes are described in Chinese, and the book is positioned as the first systematic Chinese guide to Hermes. For a team whose working language is English, the value proposition largely disappears. The README itself is bilingual only in fragments, and the substantive content is not translated.

The third limitation is scope. The guide covers monetisation, platform integration and enterprise deployment alongside architecture. That breadth is the point, but it means no single chapter is likely to be a complete reference for a specialist task. A developer who wants the MCP protocol details will read volume 11 and little else; a founder reading volume 14 gets commercial framing rather than engineering depth. The four routes acknowledge this by pruning, but pruning is not the same as depth.

How It Differs from Reading the Hermes Documentation Directly

The obvious alternative is the upstream Hermes documentation itself, which the README identifies as the source of the fragmentation problem: it is spread across multiple repositories and platforms. The difference in approach is curation versus authority. Upstream documentation is authoritative and current by construction, but a reader has to assemble a mental model from scattered pages. This guide is a single ordered narrative with four entry points, at the cost of being a snapshot with no version tracking.

A second alternative is the OpenClaw material that the guide says Hermes builds on. Volume 13 is explicitly a comparison and migration chapter, which means the guide positions itself as a bridge for readers already invested in OpenClaw rather than as a replacement for OpenClaw's own docs. If you are not migrating from OpenClaw and you do not read Chinese, neither the bridge nor the guide applies to you, and the upstream Hermes repositories are the only sensible starting point.

A third alternative, for readers who want a general agent-framework comparison, is volume 02, which the README describes as covering 2023 to 2026 development history and a comparison of mainstream frameworks. That chapter competes with any number of independent surveys, and its value depends on whether its 12-dimension comparison is kept current, which the repository gives no mechanism for.

Who Should Open This Repository, and What to Check First

The guide is a reasonable fit for a Chinese-reading developer or technical decision maker who has decided to evaluate Hermes and wants a structured on-ramp before committing time to the upstream documentation. The four routes are genuinely useful as a reading plan, and the split licence is clearly stated, which lets a reader know immediately whether their intended use is permitted.

It is a poor fit for anyone who needs English material, for anyone who wants a versioned reference they can cite against a specific Hermes release, and for anyone who wants to redistribute the prose commercially. It is also a poor fit for a reader who wants to evaluate Hermes on its own merits without the OpenClaw lineage framing, since a full volume is devoted to that comparison.

What to verify first, concretely: open 05-安装部署全攻略.md and confirm that its install steps match the Hermes version you intend to run, since the README does not reproduce those commands and the repository has no releases to pin against. Then check the licence boundary against your intended use, remembering that CC BY-NC-SA 4.0 covers the book content and MIT covers only the sample code. If you plan to build the PDF, read generate_pdf.py before running it, because the README documents only the two-command example and nothing about the script's behaviour.

Editorial conclusion

Adopt this guide if you read Chinese and want a structured path into Hermes Agent, particularly the four reading routes for beginners, developers, OpenClaw migrants and people evaluating commercial use. Do not adopt it if you need an English reference, or if you want to redistribute the prose commercially: the CC BY-NC-SA 4.0 terms on the book content rule that out, while the MIT terms cover only the sample code. Before relying on any chapter, open the specific Markdown file you intend to follow and check that its commands and config keys still match the Hermes release you have installed, because the repository carries no releases and no changelog.

Official sources

  1. Issues
  2. jwangkun/hermes-agent-guide on GitHub
  3. README
Community notes

Community notes