Learn the 21 core patterns used to build reliable AI agent systems before writing any code
Understand how to chain AI prompts together so complex tasks are broken into manageable steps
Study how multi-agent systems coordinate so you can design your own AI workflows
Read in Chinese the standard reference for AI agent architecture originally published by Google
Pure reading resource, no code to run. Browse online at adp.xindoo.xyz or clone the repo to read locally. Original English files are included alongside Chinese translations.
This repository is a Chinese translation of a book called "Agentic Design Patterns," originally published by Google. The book explains how to build AI agent systems, and this project makes that material available to Chinese-speaking readers. All 32 chapters and appendices have been translated and are hosted as a browsable website at adp.xindoo.xyz. An AI agent, in the context of this book, is a program that uses a language model to take actions, make decisions, and complete tasks with some level of independence. The book walks through 21 core patterns that developers use when building these systems. Early chapters cover foundational approaches such as prompt chaining (breaking a task into a sequence of steps), routing (directing requests to different handlers based on their type), and parallelization (running multiple model calls at the same time). Later chapters address more advanced concepts including how agents reflect on and revise their own outputs, how they use external tools, how multiple agents can coordinate with each other, and how they manage memory across a session. The final chapters cover practical concerns: how to retrieve information from a knowledge base (a technique often called RAG), how agents communicate with each other using a protocol called A2A, how to set goals and monitor progress, how to handle errors, and how to keep a human in the loop when needed. There are also appendices covering advanced prompting techniques, a survey of agent frameworks, agents on the command line, and coding-specific agents. The repository stores the translated Markdown files alongside the original English versions for reference. Readers can browse the content online or clone the repository to read locally. The project is open to contributions, primarily focused on reviewing and improving the quality of the existing translations rather than adding new content.
← xindoo on gitmyhub — every repo by this author, as a profile.
Verify against the repo before relying on details.