// the find
datawhalechina/hello-agents
📚 《从零开始构建智能体》——从零开始的智能体原理与实践教程
A Chinese-language tutorial series teaching LLM agent development from first principles, covering everything from ReAct patterns to multi-agent systems and agentic RL. It's aimed at developers with basic Python who want to build AI-native agents rather than drag-and-drop workflow tools like Dify or n8n. At 59k stars, it's clearly hit a nerve in the Chinese ML community.
The curriculum progression is well-structured — it builds from theory to implementing classic patterns (ReAct, Plan-and-Solve, Reflection) to building your own framework from scratch, which is the right order. It covers MCP, A2A, and ANP protocols in chapter 10, which is timely given how fast that space is moving. The custom HelloAgents framework (built on raw OpenAI API) gives learners something to extend rather than just consuming existing libraries. Community contributions are actively curated into extra chapters on real gaps like GUI agents, web agents, and context engineering.
It's entirely in Chinese with an English README stub, so the audience is effectively limited unless you read Mandarin. The co-creation projects folder is a grab-bag of student work with wildly inconsistent quality — some projects have no tests, hardcoded credentials in examples, or copy `.env` files committed to the repo. The agentic RL chapter (SFT to GRPO) is ambitious but a single chapter cannot do that topic justice; someone who reads it thinking they understand LLM training will be in for a surprise. No CI, no automated testing of the code examples, so there's no guarantee the Jupyter notebooks in older chapters still run against current API versions.