ranxi2001/zero2Agent

What it is

zero2Agent is a free, open-source tutorial repository (written in Chinese) for programmers who want to learn how to build their own AI "agents" — software systems that use large language models to reason, call tools, and complete tasks.

The project is not a quick demo or a list of copy-paste prompts. Instead, it's a systematic course that walks through the engineering principles behind agents, breaks down real frameworks, hands you working code, and finishes with a complete production project plus interview preparation. Nothing in the tutorial requires prior Agent experience, but it does assume you already know how to code and are comfortable with AI tools.

Who it's for (and who it's not)

Built for:

• Developers who know deep learning but have never built an LLM application
• Python programmers who want to round out their Agent engineering skills
• People who use tools like Claude Code or Cursor and want to understand how they actually work under the hood
• Students preparing for Agent-focused technical interviews or internships
• Anyone who wants to deploy an Agent in production, not just run a demo

Not for:

• Researchers studying model training, alignment, or inference optimization
• Pure academic literature reviews on Agents
• People looking for quick "build an Agent in minutes" shortcuts

What's inside

The repository currently contains 133 articles across 12 completed modules (one more module is still being updated). Here's the breakdown:

Module	Articles	What it covers
Agent Basic	17	Core concepts: what an Agent is vs. a fixed workflow, model APIs, tool calling, context, memory, planning, loops, and infrastructure
LangGraph	7	Building agents as graphs: state, nodes, conditional branches, parallel execution, prompt chaining, LLM integration
SDK Frameworks	4	The official SDKs from OpenAI, Google (genai), and Anthropic (Claude) — plus a head-to-head comparison
Framework Survey	13	Deep dives into 13 popular Agent frameworks (AgentScope, Mastra, Semantic Kernel, Eino, LangChain, AutoGen, Vercel AI SDK, and more)
OpenClaw Agent	9	Build an Agent framework from a 60-line core: follow the progression from simple workflow → chatbot → full agent with tools
Claude Code	12	A 12-lesson course hand-writing a complete coding Agent: loop → tools → subagents → teams → isolated worktrees
DeepSeek Harness	13	Design-philosophy deep-dive into an Agent runtime where every component (model, tools, filesystem, UI, even the loop itself) is replaceable via plugins
Agent Training	7	Practical training: SFT, RL, GRPO/PPO, data mix, simulation sandboxes, evaluation, deployment
Agent Application	7 (in progress)	Real-world application: vibe coding, workflows, evaluation, cloud bots
Pi Coding Agent	8	Reconstruct a production terminal Agent (Pi) into a resumable, embeddable Agent runtime
Codex CLI	7	OpenAI's terminal Agent: its Rust architecture, config, sandboxing, MCP integration
Interview Prep	17	Real interview questions from major Chinese tech companies — each with a "beginner answer" vs. "expert answer" comparison
Final Project	12	OfferPilot: a complete interview-diagnosis Agent built from scratch with a hand-written 10-layer "harness" (no LangChain or LangGraph dependency)

The free "green book"

Alongside the online course, the project publishes a free PDF called 《Agent 面试 800 问》 ("800 Agent Interview Questions") — currently at version 1.3.0. It's a condensed, interview-day cheat sheet covering 828 questions across 17 major topic areas, drawn from real interview experiences at companies like Ant, Alibaba, ByteDance, Tencent, Ctrip, and Baidu. The companion online question bank has been updated to 844 questions. The book is licensed under CC BY-NC-SA 4.0 (free to share with attribution, non-commercial).

How it's built

The repository is a Jekyll-based static site designed to be hosted on GitHub Pages. You can read it online at onefly.top/zero2Agent, or clone the repo and run it locally:

git clone https://github.com/ranxi2001/zero2Agent
cd zero2Agent
# Install Jekyll (requires Ruby)
gem install bundler jekyll
bundle install
# Local preview
bundle exec jekyll serve

Credits and licensing

The tutorial openly credits its sources: it draws on and adapts content from several existing open-source projects (like the Claude Code course by shareAI-lab, the Learn-OpenClaw project, and the official DeepSeek Harness repository), always with attribution and, in some cases, original rewrites. The repository itself is licensed under the MIT license — meaning you're free to use, modify, and distribute it, even commercially.

In one sentence

If you're a programmer who wants to move from using AI tools to building genuine agent systems — and to do it properly, with production-grade architecture and interview-grade depth — zero2Agent walks you through the entire journey, one well-structured article at a time.