ai-agent-book

An open source book that walks from Agent = LLM + context + tools to production engineering, with the full text and companion experiments published openly.

The book hangs on one equation: Agent = LLM + context + tools. Around that formula it builds ten chapters that move from design principles to engineering practice. The entire text, the figures, and 95 companion experiments are open source, so the material can be run rather than just read. Chinese is the original language, and the README lists a dozen more language versions, most of them community translations.

The ten chapters run from basics up to production concerns. The 95 experiments split between local projects and external reproduction tracks, with each chapter pointing at its own code. A quick view of the first and tenth chapters gives a flavor of the range, and a study guide covers learning paths, difficulty levels, and practice advice. Running the experiments means configuring provider keys first; some experiments support local Ollama only when the experiment README or CLI explicitly says so.

The ebook is available for offline download as PDF and EPUB, always built from the latest main branch. Chinese is the original; the other language editions are community translations that may lag behind it. Online reading supports multi language switching, chapter folding, full text search, and direct links to the companion experiments, and the online build rebuilds automatically after pushes to main. Building the compiled version yourself requires pandoc, xelatex, the ElegantBook document class, and the related fonts.

The Chinese text source lives in a book directory. English, Spanish, Indonesian, and the other translations sit in their own directories, which is why they can fall behind the original. Chapter files run from the introduction through chapter 10 and an afterword. The contribution section lists book content improvements, code and bug fixes, new practice projects, diagram improvements, and new translations as the main ways in.

The repository is a rare thing: a book with its entire text, figures, and experiments published openly. Anyone can read it in the browser, download it, or rebuild it from source.

• Source 1 · Project repository
• Source 2 · README