Preface

Picture a first-year student facing a brand-new course: textbooks, lecture notes, past exam papers and mistake logs scattered across different folders. They do not know where to start, and they cannot tell which topics they are actually weak in. This is the real predicament of countless learners — and agents offer a new way out: a study companion that understands you and prepares alongside you.

This book grew out of a more general question: how do you turn an idea into a working agent application using a substrate like DeepSeek Harness? Throughout, building comes first — get the harness running, wire up a model, write a plugin, then build the agent. The running example, the exam-prep assistant ShuTongBuddy, is the natural product of that path: it starts as one plugin package and grows into a complete web application.

The material was compiled from the DeepSeek Harness repository’s README, user guide, architecture documents and cookbook. The DeepSeek AI team open-sourced the project under the MIT licence, providing a solid technical foundation for this book — my sincere thanks. I also thank the editorial team of Large Model Agent Application Development for their professional support in refining the manuscript.

Because DeepSeek Harness is still in developer preview and iterates quickly, a few details should be checked against the official documentation; such passages are marked 【待核】 in the text. If you find mistakes, please let me know.

Guo Tao Email: guotao3s@163.com