datawhalechina/hello-agents

What it solves

Building an AI agent today is full of fragmented tutorials and framework-specific guides, but there is almost no systematic, practice-first learning path that takes you from the core concepts to actually coding your own agent. This project fills that gap with a complete, open-source curriculum designed to turn a user of large language models into a builder of agent systems.

How it works

The tutorial is organized into five parts, progressing from foundations to advanced skills. Part one explains what agents are, their history, and the basics of large language models. Part two walks you through classic patterns like ReAct and Plan-and-Solve, low-code platforms such as Coze and Dify, mainstream frameworks like LangGraph, and finally guides you in building your own agent framework from scratch using the OpenAI native API. Part three covers memory and retrieval, context engineering, communication protocols (MCP, A2A, ANP), training an agent with reinforcement learning (from SFT to GRPO), and performance evaluation. Part four applies everything in real projects: a travel assistant, an automated deep-research agent, and a simulated town. Part five ends with a graduation project where you build your own multi-agent application.

All code is provided in the repository's code folder, and readers are encouraged to run, debug, and modify it. The project also includes community-contributed extras, like interview questions, environment setup guides, and additional practical essays.

Who it’s for

The project is aimed at AI developers, software engineers, students, and self-learners who already have basic Python skills and a conceptual understanding of calling an LLM via an API. You do not need a deep background in algorithms or model training, since the focus is on practical application and construction.

Highlights

• Completely free and open source, with a Chinese-language curriculum and an online reading version
• Covers the full journey from agent theory and classic paradigms to advanced techniques like memory, context engineering, protocols, and agentic RL
• Includes practical work: build a travel assistant, a deep-research agent, and a simulated town
• Teaches you to create your own agent framework from zero using the OpenAI native API
• Offers community contributions such as interview questions, setup guides, and hands-on tutorials
• Provides a downloadable PDF with a watermark to prevent re-selling