This repository offers a comprehensive, 36-chapter full-stack curriculum for mastering AI Agents, designed for developers seeking systematic knowledge and interview preparation. It provides over 22,000 lines of runnable Python code across 60+ examples, covering foundational theory, advanced engineering practices, and cutting-edge research, enabling users to build and deploy sophisticated AI agent systems.

How It Works

The curriculum is structured into 7 progressive layers, with each chapter presented as an independently runnable Python file that doubles as a detailed lecture and executable code. This "lecture-as-code" approach integrates theoretical concepts with practical implementation, covering topics from basic ReAct loops and function calling to complex protocols like MCP and A2A, RAG, DSPy, and production observability. The layered design allows for a gradual build-up of knowledge, from fundamental agent components to advanced architectural patterns and expert-level techniques.

Quick Start & Requirements

Clone the repository (git clone https://github.com/Callous-0923/agent-study.git). Install dependencies by running chapter_00_overview/00_course_overview.py with install = True. Chapters 1-5 require API keys (e.g., OPENAI_API_KEY) configured in a .env file; alternative domestic models can be used. Notably, chapters 8-28 and beyond largely rely on Python's standard library, allowing execution without external API access. Links to official documentation are not provided within the README.

Highlighted Details

• Comprehensive Coverage: Deep dives into Tool Calling (OpenAI/Anthropic), MCP and A2A protocols, Claude Code architecture, RAG, DSPy, Agentic Workflows, Context Engineering, Streaming, Observability, and Agent Security.
• Runnable Code: Each chapter is a self-contained .py file serving as both lecture notes and executable demonstration code.
• Interview Focus: Explicitly targets job seekers with high-frequency interview questions, answer frameworks, and scoring points highlighted per chapter.
• API Key Flexibility: A significant portion of the course (Ch8-28+) can be run using only standard Python libraries, reducing setup friction.

Maintenance & Community

The project is marked as "continuously updated" with an invitation for community contributions via Issues and Pull Requests. Specific links to community channels (Discord/Slack), active maintainers, or a public roadmap are not detailed in the README.

Licensing & Compatibility

The project is released under the MIT License, permitting free use, modification, and distribution. This license generally allows for integration into commercial and closed-source projects without significant restrictions.

Limitations & Caveats

Initial chapters (1-5) necessitate API key configuration for specific LLM providers. While many chapters are self-contained, others depend on external libraries like LangChain or FastAPI, requiring installation. Some advanced topics, such as reverse-engineered architectures (Claude Code) or pre-release protocol specifications (MCP/A2A), may be subject to change or community interpretation.