What it does Hello-Agents is a free, 16-chapter Chinese tutorial that walks you from agent history and LLM fundamentals to building your own framework and training models with Agentic RL. It covers ReAct, Plan-and-Solve, and Reflection patterns, then moves into memory systems, context engineering, MCP/A2A protocols, and evaluation. The capstone: a multi-agent travel assistant, a DeepResearch clone, and a “cyber town” simulation.

The interesting bit The authors draw a sharp line between “software engineering agents” (Dify, Coze, n8n — basically LLM-backed workflows) and “AI-native agents” where the LLM actually drives decisions. The course targets the latter, and includes a from-scratch framework called HelloAgents built on raw OpenAI APIs. There’s even a chapter on post-training your travel planner from demo to production-grade model.

Key highlights

• 16 chapters, all marked complete, with community-contributed extras (interview prep, GUI agents, web agents, self-evolution)
• Hands-on framework construction alongside tutorials for AutoGen, AgentScope, and LangGraph
• Agentic RL coverage from SFT through GRPO — unusual depth for a tutorial
• PDF release with an embedded open-source watermark to deter resellers
• Hosted reading with China-optimized mirror at hello-agents.datawhale.cc

Caveats

• Entirely in Chinese; English README exists but the content does not appear translated
• The “build your own framework” chapter links to a separate repo (jjyaoao/helloagents) — unclear how tightly integrated it is

Verdict Worth bookmarking if you’re a Mandarin-speaking developer who wants to understand agents below the framework abstraction layer. Skip it if you’re looking for quick no-code recipes or English-language instruction.