Summary

This repository offers a comprehensive 15-week engineering curriculum, "100 天搞定 Agent 开发," designed to guide developers from fundamental LLM understanding to building robust, controllable Agent systems. It addresses common real-world challenges in Agent development, such as unmanageable prompts and uncontrolled tool usage, providing a structured path for engineers aiming for maintainable and scalable AI applications.

How It Works

The project follows a phased learning approach, starting with LLM principles and progressing through core Agent components like Prompt Engineering, Tool Use, RAG, Context, and Memory. It then delves into Agent architectures, patterns (ReAct, Plan-and-Execute), and advanced concepts like Reflection, Task Decomposition, and Human-in-the-Loop (HITL). The curriculum culminates in building a foundational Agent platform featuring a Gateway, Runtime, Plugin System, and multi-channel support, leveraging frameworks such as LangChain, LangGraph, and Gradio. This methodology prioritizes practical engineering and addresses architectural complexities often missed in simpler demo projects.

Quick Start & Requirements

The primary environment is Python with Jupyter Notebook. Installation involves installing uv (pip install uv), syncing dependencies (uv sync), configuring API keys in .env, and launching Jupyter Notebook (uv run jupyter notebook). Prerequisites include a Python environment and necessary API keys for LLM providers.

Highlighted Details

• Structured 15-Week Curriculum: Covers LLM fundamentals to advanced Agent platform engineering.
• Practical Engineering Focus: Addresses real-world pitfalls in prompt management, tool control, and system scalability.
• Comprehensive Agent Components: Explores Prompt Engineering, Tool Use (MCP), RAG, Context, Memory, Reflection, Task Decomposition, and HITL.
• Platform Architecture: Details building a Gateway, Runtime, Plugin System, and multi-channel support.
• Framework Integration: Hands-on examples with LangChain, LangGraph, Gradio, Agno, Mem0, Vanna, Langfuse.

Maintenance & Community

The project is presented as continuously evolving, welcoming contributions via Issues, Forks, and PRs. It encourages community feedback and corrections based on real-world practices.

Licensing & Compatibility

The provided README does not specify a software license. This absence is a critical factor for evaluating commercial use or derivative works.

Limitations & Caveats

The project is structured as an educational path and actively evolving, suggesting it may not be production-ready without further hardening. The lack of a specified license poses a significant adoption barrier, particularly for commercial applications. Setup requires familiarity with Python environments and API key management.