hermes-agent-orange-book README

The project's README from github.com/alchaincyf/hermes-agent-orange-book. As one markdown file: README.md.

English | 中文版 README

hermes-agent-orange-book Hero Animation
Animated with huashu-design skill

Hermes Agent 2.0: The Complete Guide

橙皮书 (Orange Book) Series · by HuaShu (花叔)

A hands-on guide to Hermes Agent, the open-source AI Agent framework by Nous Research — the first agent that ships with its "reins" built in, and the reins grow themselves.

This is a from-scratch rewrite. The first edition was based on Hermes v0.7.0. Two months and nine releases later, the product had grown a whole new face — a native desktop app, a full browser dashboard, 23 messaging platforms — so the book is rebuilt around v0.16.0 ("The Surface Release").

Download

Version PDF
中文版 (Chinese) PDF Download
English PDF Download

Errata

The data-collection inference in §03 ("Why Nous Built It") is wrong. Thanks to @lishaogang for pointing this out in issue #7.

Under "Motive one," the book claims that "Hermes Agent isn't just a product, it's also a massive-scale harvester that collects real-world data for Nous's own models," and draws a chain from "users' real usage → tool-call trajectories → compressed into training data → next-gen models." That chain breaks at its very first link:

In other words, your usage trajectories stay on your own machine; there is no channel feeding them back to Nous. The things the book cites — batch_runner.py, trajectory_compressor.py, the "Research-ready" section — are all real, but they are research infrastructure for Nous and researchers to batch-generate trajectories in their own environments, not a pipeline collecting data from users. I stretched "they have trajectory-generating infrastructure" into "they are harvesting your trajectories," and that step went too far. The original text did flag it as "my inference, not an official Nous statement" — but an inference directly contradicted by the official docs is still wrong, flag or no flag.

Accordingly, the "data flywheel" argument at the end of §03 doesn't hold either. Of the motives the book attributes to Nous, "monetizing through Portal" and "poaching OpenClaw's users" still stand; "users feeding it training data" should be struck.

The PDFs are build artifacts and have not been regenerated yet; this erratum stands here until the next revision corrects the body text. For product facts, trust the official docs.

What This Book Covers

Hermes Agent is an open-source AI Agent framework from Nous Research. Unlike OpenClaw and Claude Code, it takes a fundamentally different approach: a built-in self-improving learning loop, a three-layer memory system, automatic Skill creation and evolution, and — new this cycle — a persistent multi-agent Kanban platform and an honest, OS-level security model.

If you've read the "Harness Engineering" 橙皮书, Hermes is the productization of those five components (instructions / constraints / feedback / memory / orchestration).

21 sections across 6 parts:

Part Content Sections
1. What It Is Meet Hermes, one brain many faces, why Nous built it §01-03
2. The Reins Grow Themselves Three engines of self-improvement, Curator, what not to learn §04-06
3. How It Remembers You Three-layer memory, session search, Skills, orchestrating agents §07-10
4. Connecting Everything 64 tools, MCP, 23 platforms, three surfaces §11-14
5. Multi-Agent & Orchestration From delegate_task to the Kanban platform, collaboration patterns §15-17
6. Deployment, Security & Boundaries Deploy, the OS boundary, Promptware defense, how far it can go §18-21

Who Is This For

橙皮书 (Orange Book) Series

This is part of the 橙皮书 series — free, practical guides on AI tools. Other titles include Claude Code, Harness Engineering, OpenClaw, and more.

All books free to download: huasheng.ai/orange-books

About the Author

HuaShu (花叔) · AI Native Coder · Indie Developer

An AI content creator with 300K+ followers across platforms. Built all products (including an App Store #1 Paid iOS app) entirely with AI tools — never wrote a line of code manually. Open-sourced Nuwa.skill, huashu-design, and more.

Version

Changes in 2.0

📚 Read Online on WorkBuddy

This book is also available on WorkBuddy's knowledge base, one page per chapter — or just drop the link into your agent as context.

https://www.workbuddy.cn/space/d/qAVr77wMMaD1AmkPtmL501

All orange books: https://www.workbuddy.cn/space/d/YcllWXknAUoMk6lFSWdfbI

License

MIT License — free to use, copy, modify, and distribute, including for commercial use. Attribution appreciated but not required.

hermes-agent-orange-book (alchaincyf/hermes-agent-orange-book)

Searching Database…

5K ·GitHub
Hermes-agent-orange-book
What is this?This is a free ebook (the Orange Book series) that explains Hermes Agent, an open-source AI Agent framework from Nous Research, covering its self-improvement loop, memory system, tools, and multi-agent orchestration.
Is this the Agent framework itself or just a book about it?It is a book (PDF, available in Chinese and English) that documents and explains Hermes Agent; the repo does not contain the Hermes Agent software itself.
How do I read this book?Download the PDF (Chinese or English) directly from the README's download table, or read it online chapter-by-chapter on WorkBuddy's knowledge base.
Do I need to know Hermes Agent already before reading?No, the book starts from the basics ('What It Is') and builds up, though it's aimed mainly at people who've used tools like Claude Code, OpenClaw, or Cursor.
Is everything in the book fully accurate?No — the author published a correction: the §03 claim that Hermes harvests user trajectories to train Nous's models is wrong and contradicted by Hermes' official no-telemetry policy, though the PDFs haven't been regenerated yet to fix this.
Who is this book for?Developers familiar with Claude Code, OpenClaw, or Cursor; AI power users who prefer a desktop app over the command line; and anyone curious how Harness Engineering concepts become a real self-improving product.
Does Hermes Agent send my usage data back to Nous?No — per Hermes' own FAQ, API calls go only to the LLM provider you configure, there is no telemetry or usage collection, and conversations/memory/skills are stored locally.
How is the book organized?It has 21 sections across 6 parts: what Hermes is, its self-improvement engines, its three-layer memory system, its 64 tools/MCP/23 messaging platforms, multi-agent orchestration via a Kanban platform, and deployment/security boundaries.
What makes Hermes Agent different from Claude Code or OpenClaw?The book says Hermes has a built-in self-improving learning loop, automatic Skill creation and evolution, a persistent multi-agent Kanban platform, and an OS-level security model, which it frames as the productization of 'Harness Engineering' concepts.
Is this the first or a revised edition?It's the second edition (v260607), a full from-scratch rewrite based on Hermes Agent v0.16.0, replacing the first edition (v260408) which covered the older v0.7.0.
What are similar projects?zhijian-ai-bluebook-workbuddy-harness, claude-code-book, vibe-coding-cn, awesome-harness-engineering, harness-engineering, deepagents, Benzi, deepseek-harness-orange-book
Which topics is it in?harness-engineering, ebook, openclaw, harness, book, multi-agent, learning-resources, documentation, local-first, agent, privacy, claude-code
Creator
alchaincyf
Size
39.6 MB
Stars
5K
Updated
CommitAug 23
Age
6mo
Commits · 90 Days