Open-source AI Agent engineering book with 109 runnable experiments

What they did

The author (bojieli) published a full book, 深入理解 AI Agent：设计原理与工程实践, built around the formula Agent = LLM + context + tools. The repo ships the complete manuscript, figures, and 109 companion experiments, including local projects and an external-reproduction track. The book runs 10 chapters from fundamentals to production, and version 2.0 restructured the material: the old chapter 4 async-interaction section merged with chapter 9's multimodal-agent content into a new chapter 6 on interaction and expanding observation/action spaces, pushing evaluation, post-training, and continual evolution to chapters 7-9. Distribution is handled through GitHub Releases with always-latest PDF and EPUB builds, plus an online Astro reader with language switching, chapter folding, highlighting, and notes that auto-updates on each push. Community translators contributed 15 language editions (English, Spanish, Indonesian, Traditional Chinese, Russian, Tamil, Vietnamese, Japanese, Turkish, Korean, Hungarian, Hebrew, Brazilian Portuguese, Arabic), each with its own PDF/EPUB. A sister volume, 深入理解 AI Infra, covers the infrastructure layer underneath model training and inference.

How it went

A polished, actively maintained open-source book that hit GitHub Trending Project of the Day, with a stable release pipeline, a live reading site, and a broad community translation program. The 109 experiments give readers a concrete path from theory to hands-on practice, and the 2.0 restructure tightened the chapter arc. Limits are mostly around the experiments themselves: they are reader-run, so results depend on the reader's own model access, API keys, and environment.

Worth knowing

The primary text is Chinese; other languages are community translations and may lag the source. Experiment reproduction depends on external model APIs and local setup that the README does not fully enumerate. The 2.0 restructure means older PDFs and third-party notes reference outdated chapter numbers, so pin to a release if you cite chapters.

Wiring

Start from the repo README, which links the latest PDF/EPUB under Releases and the online Astro reader. Choose a language edition, then download the offline build for best layout or use the web reader for notes and highlighting. For the experiments, clone the repo, locate the experiment directory referenced in the relevant chapter, and follow its local setup; experiments that need model access require your own API credentials. The sister repo ai-infra-book is a separate clone covering the infrastructure layer.

Try it yourself

Read the ai-agent-book repo README and the table of contents. Pick three experiments that map to the Agent = LLM + context + tools formula, list each one's prerequisites (model access, API keys, runtime), and draft a step-by-step plan to run them locally in order, noting where I'll need to substitute my own model provider.