Hello-Agents: AI Agent Development Tutorial
/hello-agents
Hello-Agents is a 16-chapter curriculum from the Datawhale community that walks from agent theory through building and training your own multi-agent system, and every chapter is marked complete in its table of contents. Reach for it if you read Chinese comfortably and want a structured path from ReAct to Agentic RL; skip it if you need English-first material or a single framework's docs instead of a full course.
Introducing the Hello-Agents Curriculum
Hello-Agents is a free curriculum from the Datawhale community that teaches AI agent development from first principles through to a working multi-agent build. It's organized into 16 chapters across five parts, readable online through GitHub Pages or a mainland-China mirror site, with an optional PDF and a companion code framework called HelloAgents.
Addressing the AI Agent Learning Gap
Hello-Agents responds to a fairly specific split the maintainers describe in how people currently learn to build agents. One school treats an agent as a flowchart wrapped around an LLM call, the pattern used by tools like Dify, Coze, and n8n, where the logic is process-driven and the model just handles one step of data processing. The other school builds agents actually driven by the model's own reasoning loop. The README argues that as attention in 2025 shifted from training larger foundation models toward building smarter agent applications, resources that teach this second, AI-native style systematically, from theory through a working multi-agent build, were scarce, and that gap is what the project set out to close.
What You Will Learn
- ✓Core theory across 3 chapters: agent definitions, types, and paradigms; the history of agents from symbolic AI to LLM-driven systems; and LLM fundamentals like Transformer architecture, prompting, and the limits of mainstream models.
- ✓Hands-on paradigm building in chapter 4: implement ReAct, Plan-and-Solve, and Reflection yourself.
- ✓Low-code platform walkthroughs in chapter 5: Coze, Dify, and n8n.
- ✓Framework practice in chapter 6 across AutoGen, AgentScope, and LangGraph, plus building your own agent framework from scratch in chapter 7.
- ✓Advanced systems in chapters 8 through 12: memory and retrieval (RAG), context engineering, agent communication protocols (MCP, A2A, ANP), and formal agent performance evaluation.
- ✓A dedicated Agentic RL chapter (11) that walks through the full SFT-to-GRPO pipeline for training an LLM.
- ✓A companion framework called HelloAgents, hosted separately at github.com/jjyaoao/helloagents and built from scratch on the raw OpenAI API, used as the codebase for the hands-on chapters.
- ✓A Community Blog of contributor-written Extra-Chapters covering things like interview prep, Agent Skills versus MCP, and GUI agents.
Practical Agent Projects
- •Building an intelligent travel assistant that combines MCP with multi-agent collaboration, in chapter 13.
- •Reproducing a DeepResearch-style automated research agent in chapter 14.
- •Building a simulated "cyber town" where multiple agents interact in a game-like environment to model social dynamics, in chapter 15.
- •Shipping a capstone project in chapter 16, where you design and complete your own full multi-agent application.
- •Prepping for agent-related job interviews using the interview question set contributed to the community Extra-Chapter collection.
Strengths
- ✓Free and complete: Datawhale distributes it at no cost, and all 16 chapters across five parts are marked done in the repo's own table of contents.
- ✓Opinionated about what an agent actually is, teaching the AI-native, model-driven pattern instead of only flow-based, low-code building.
- ✓Theory pairs with runnable code: a code folder ships alongside the text, and the README pushes readers to actually run, debug, and modify every example rather than just read it.
- ✓Goes further than most intro material, into Agentic RL (SFT to GRPO) and formal agent evaluation, topics that beginner-focused tutorials usually skip.
- ✓An active community layer: a growing set of contributor Extra-Chapters adds interview prep, protocol comparisons, and topic deep-dives beyond the core 16 chapters.
Considerations Before Starting
- △The README shows an English/Chinese toggle at the top, but the project introduction, chapter table, and community content are all in Chinese, so full English translation of every chapter isn't confirmed from what's given here.
- △The content license is Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International, which blocks commercial reuse of the material, fine for personal study but restrictive if you wanted to repackage it.
- △There's no version number or dated changelog for the curriculum itself; the README's "next steps" section is a running list of plans (more videos, a future book on training an agent from scratch) rather than a release schedule.
- △This is instructional content, not a runnable package: there's no install command or API to call, just chapters, code samples, and a downloadable PDF.
- △HelloAgents, the companion framework built during the hands-on chapters, lives in a separate repository (jjyaoao/helloagents) from the curriculum, so its own maturity and versioning aren't covered by these facts.
Exploring Agent Frameworks and Platforms
Common Questions
Hello-Agents is completely free. The Datawhale community distributes all chapters, the PDF, and the accompanying code at no cost under an open license.
Hello-Agents expects basic Python programming ability and a conceptual grasp of large language models, like knowing how to call one through an API. A deep algorithm or model-training background isn't required to start.
Hello-Agents welcomes contributions through GitHub pull requests, whether that's a bug fix, a chapter improvement, or a new write-up submitted to the community Extra-Chapter collection.
Hello-Agents content is released under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license, so reuse is free but commercial use isn't permitted without separate permission.
Hello-Agents doesn't have a full video course yet; the README lists video lectures as an upcoming addition, planned to go deeper into practical build steps than the written chapters.
Hello-Agents dedicates Chapter 5 to low-code agent platforms, walking through Coze, Dify, and n8n before moving into code-first frameworks the following chapter.
Who should try it — and who should skip
Try Hello-Agents if you're a developer or student with basic Python who wants a structured, chapter-by-chapter path from agent theory into training and evaluating your own multi-agent system, and you don't mind a Chinese-first curriculum. Skip it if you need English-only material, want a single framework's reference docs rather than a full course, or need something production-ready this week instead of a multi-week study plan.
Related repositories
Still deciding about hello-agents?
One click hands the question to an AI along with this page — see what it says about hello-agents.