https://github.com/alienzhou/zero2agent
A hands-on course on building an AI agent from scratch — code included, with a full step-by-step build log, based on real product experience.
https://github.com/alienzhou/zero2agent
Last synced: 4 months ago
JSON representation
A hands-on course on building an AI agent from scratch — code included, with a full step-by-step build log, based on real product experience.
- Host: GitHub
- URL: https://github.com/alienzhou/zero2agent
- Owner: alienzhou
- Created: 2026-03-10T01:58:49.000Z (4 months ago)
- Default Branch: main
- Last Pushed: 2026-03-11T16:24:10.000Z (4 months ago)
- Last Synced: 2026-03-11T21:52:24.409Z (4 months ago)
- Language: Python
- Size: 482 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.en.md
- Changelog: CHANGELOG.md
- Agents: AGENTS.md
Awesome Lists containing this project
README
Zero2Agent
🚀 A Hands-on Course for Building Production-Grade AI Agents from Scratch
Dive into engineering details, build your first Coding Agent
---
## 🎯 About
> There's plenty of Agent-related content out there—papers, tutorials, products, open-source projects—but few teach you how to "build a production-grade Agent from scratch." This repository is such a **teaching case project**.
- **Starting from the first line of code**, step by step building a production-grade Coding Agent similar to Claude Code / Codex
- **Completely open and transparent**, including requirements analysis, design decisions, pitfalls, and detours
- **Recording the real process of collaborating with AI**, Vibe Coding / Agentic Engineering will be applied throughout development
---
## ✨ How Is This Different?
| Dimension | Other Courses | Open-Source Products | Zero2Agent |
| ---------------------- | -------------------------------------- | ------------------------------------- | --------------------------------------------------------------------- |
| **Engineering Practice** | Concept-focused, demo-level code | Only final code, no process | Deep dive into real engineering problems, distilled from actual Agent development experience |
| **Production-Grade** | Basic features and cases | Complete but complex, hard to learn | Features curated from real products, designed as hands-on material |
| **Step-by-Step** | Chapter-based, large gaps, not detailed | Code changes too complex to follow | Each iteration can be followed independently, right-sized, progressive |
---
## 🎓 Is This for You?
If you:
- 🌱 **Want to get started with LLM application development**, but don't know where to begin
- 🤖 **Want to learn AI Agent development**, but find blog posts too abstract and frameworks too black-box
- 🛠️ **Want to see what real AI-assisted development looks like**, not the "done in 10 minutes" marketing stories
- 📚 **Prefer learning by doing**, rather than just reading theory
Then this teaching project is for you.
---
## 📦 What You'll Get
### 📖 See the Complete/Real AI Agent Building Process
Content distilled from production projects as teaching cases, not purely a Toy Project, but based on real development:
- Starting from actual problems/requirements
- Including requirements discussion records (why we did this, not that), design documents (spec for each iteration)
- Accompanying code implementation
- Retrospective notes (what went right, what went wrong)
### 🤖 Learn AI-Assisted Development
This project also uses AI-assisted development throughout, itself a journey of Vibe Coding/Agentic Engineering. You can see:
- What actual coding conversations and prompts with AI look like
- Practice of SSD development and other patterns
- How to use AI to do more
### 🔧 Pressure-Free Follow-Along Mode
Every iteration has a Git tag, you can:
```bash
git checkout E01-S001-react-basic # Jump to any iteration
```
Fork it and get hands-on—that's the best way to learn. Don't worry, you can enter at **any time, from any progress point** (git tag) to follow along, or pick the topics that interest you.
---
## 🗺️ Course Roadmap
The course content is organized into four layers:
- **README / Homepage**: quickly understand the project and where to start
- **Roadmap Overview**: see the full learning map first
- **Epic Page**: understand why a stage exists
- **Story Page**: enter a concrete topic and learn how to read the implementation
### Current Roadmap
| Epic | Goal | Status |
| ---- | ---- | ---- |
| [Epic 1: Read / Search](./specs/E01-read-and-search/README.md) | Build a safe, explainable minimal read-only loop for the Agent | 🚧 In Progress |
| Epic 2: Act / Modify / Execute | Move the Agent from "can inspect" to "can take action" | 📝 Planned |
| Epic 3: Core Capabilities and Productization | Move the Agent from a demo toward a usable product shape | 📝 Planned |
| Epic 4: Robustness and Context Management | Handle failures, long context, and complex runtime situations | 📝 Planned |
| Epic 5: Extensibility | Add AGENTS, Skills, MCP, Hooks, and other extension capabilities | 📝 Planned |
**Recommended starting points**:
- Start with the [Course Roadmap Overview](./docs/roadmap/README.md)
- Then go to [Epic 1: Read / Search](./specs/E01-read-and-search/README.md)
- If you want to see a full Story example directly, start from [E1-S1: Bootstrapping the Minimal Read-Only Loop](./specs/E01-read-and-search/S001-react-basic/README.md)
---
## ⚠️ What This Is NOT
This is not an Agent product intended for direct production use, it's more of a "teaching tool."
If you're looking for an out-of-the-box AI Agent, try Claude Code, Cursor, Codex, or projects like Open Code, PI.
This is a **learning resource**, not a pure tool.
---
## 📁 Project Structure
```
zero2agent/
├── packages/ # Code
│ ├── core/ # Agent core logic
│ ├── tui/ # CLI interface
│ └── shared/ # Shared code
├── specs/ # Design documents
├── retros/ # Retrospective notes
├── .vibecoding/ # AI collaboration records
├── .discuss/ # Requirements discussion records
└── CHANGELOG.md # Iteration log
```
---
## 📈 Iteration Progress
| Iteration | Content | Status |
| --------- | ----------------- | --------- |
| E01-S001 | Basic Agent Loop | ✅ Done |
👉 See full iteration details and learning guides:
- [Course Roadmap](./docs/roadmap/README.md)
- [CHANGELOG.md](./CHANGELOG.md)
---
## 🚀 Quick Start
```bash
git clone git@github.com:alienzhou/zero2agent.git
cd zero2agent
pnpm install && pnpm build
pnpm --filter @zero2agent/tui start
```
Requirements: Node.js >= 22.0.0, pnpm >= 9.0.0
---
## 📄 License
MIT