GitHub radar
Hello-Agents: a free course on building an AI agent from scratch
This is a free 16-chapter book on how agents actually work inside — from a basic reasoning loop to full frameworks. After it, you'll understand what your agent does at each step, not just how to prompt it.
An open guide from Datawhale on building AI agents genuinely driven by the model's own reasoning — not a low-code builder like Dify or n8n.
When it helps
Useful if you already work with coding or chat agents and want to see what happens inside the reasoning-action loop. Skip it if you just need a ready-made tool — this is a book, not a library.
Pros
- Free, with working code at every step, not just theory
- Shows how an agent decides to act and checks its own output
- Covers communication protocols between agents (MCP, A2A, ANP)
- Available in English and Chinese
Cons
- Chapters are dense and sequential — skipping risks losing context
- Requires basic Python and some experience calling a model via API
How to set it up — step by step
- 1Open the Hello-Agents repository (datawhalechina/hello-agents) on GitHub.
- 2Run a coding agent, such as Claude Code, in your project folder.
- 3Send the agent the text from the block below.
- 4If the agent asks for a language-model API key, give it one.
- 5Check that the agent showed the output of each reasoning-action-observation step.
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Open the Hello-Agents repo (datawhalechina/hello-agents), read chapter 4 on classic agent paradigms (ReAct, Plan-and-Solve), and go through the example code in that chapter's code folder. Run the example and show me what each step of the reasoning-action-observation loop prints. Ask me for a language model API key if the example needs one.
▌ More finds