Hello-Agents is an open-source tutorial on building AI agents, published in Chinese by the Datawhale community under the title 《从零开始构建智能体》 and cited in English as Hello-Agents: Building an AI Agent from Scratch. It is distributed free through the project’s repository, an online reading site and downloadable PDF releases, and is licensed CC BY-NC-SA 4.0.
The premise the introduction states is that if 2024 was the year of competing foundation models, 2025 opened the “year of the Agent” — the technical focus shifting from training larger base models to building more capable agent applications — while systematic, practice-heavy tutorials remained extremely scarce. The project positions itself against that gap.
Its scope is narrowed by a distinction the introduction draws between two kinds of agent building. One is the software-engineering kind, exemplified by Dify, Coze and n8n, which it characterizes as process-driven software development with the LLM acting as a data-processing backend. The other is the AI-native kind, genuinely driven by the model. The tutorial states that it is about the second, and describes its aim as taking the reader through the framework surface to the core principles, architecture and classic paradigms, and finally to building their own multi-agent application.
Contents
The book runs to sixteen chapters in five parts, and the repository marks all of them complete. Part one covers agents and language model foundations: what an agent is, its types and paradigms; a history running from symbolic AI to LLM-driven agents; and a chapter on Transformers, prompting, the major LLMs and their limits. Part two is the hands-on core — implementing ReAct, Plan-and-Solve and Reflection by hand, then working with low-code platforms (Coze, Dify, n8n), then mainstream frameworks (AutoGen, AgentScope, LangGraph), and finally building an agent framework from zero. Part three extends into memory and retrieval, Context engineering, inter-agent communication protocols including Model Context Protocol, A2A and ANP, Agentic RL from SFT through GRPO, and agent performance evaluation. Part four is three integrated projects: a travel assistant, an automated deep-research agent, and a simulated “cyber town” of agents modelling social dynamics. Part five is a capstone in which the reader builds a complete multi-agent application of their own.
Alongside the main text the repository carries a community-contributed Extra-Chapter series — interview questions for agent roles, a supplement on context engineering, a Dify walkthrough, a comparison of Agent Skills with MCP, GUI and Web agent primers, guidance on writing a good Skill, a collection of practical pitfalls from agent application development, and a chapter on agent self-evolution.
The later parts are worked through a self-built framework the project calls HelloAgents, described as built from scratch on OpenAI’s native API and since updated to v1.0.0; part three states that it uses that framework from part two rather than a third-party one.
Characters & Publication
The book is a Datawhale community project, described by the repository as entirely open-source and free: it is distributed through GitHub, an online reading site with a separate mirror for readers in China, and PDF releases. The project notes that the PDFs carry a Datawhale watermark, added to deter resellers from repackaging the free material.
The citation the repository supplies gives the year as 2025 and names seven contributors ahead of “all Hello-Agents contributors”. The repository’s own acknowledgements describe the roles behind those names — a project lead responsible for the full text, two co-initiators who contributed a chapter and the exercises respectively, several chapter contributors, and an academic advisor — and credit a further set of named contributors for the Extra-Chapter material. The repository’s forward-looking list names a follow-up work on training agents from scratch, alongside video courses and further development of the HelloAgents framework.