Annyfee/agent-craft

What it solves

Agent Craft addresses the gap between knowing AI agent concepts and being able to build them from scratch. Many learners understand theory or can call APIs, but struggle to assemble a complete, runnable agent. This project provides a structured, module-by-module learning path that takes you from basic LLM calls to deploying a full agent system.

How it works

The project is organized into 15 sequential modules (currently 01–13 are open), each with independent code, comments, and examples. You progress through:

• Fundamentals: setting up the environment, calling LLMs, and enabling Function Calling.
• Framework layer: using LangChain for prompts, chains, and memory; then RAG with vector stores (FAISS, Chroma) and rerankers; then LangGraph for stateful, debuggable workflows, including human-in-the-loop and multi-agent orchestration.
• Intelligence layer: building MCP servers and clients with FastMCP and langchain-mcp-adapters, plus exploring the Swarm pattern and Agents SDK.
• Practical projects: Streamlit for building a chat UI, with a future comprehensive project combining LangGraph, RAG, MCP, Streamlit, and Vercel.

Each module comes with a matching CSDN blog post explaining the design and key mechanisms. The entire project emphasizes code that runs locally and principles that are easy to grasp.

Who it’s for

This is for developers and learners who already know basic Python and want to understand AI agents from the inside out — not just call frameworks. If you’ve felt stuck between “knowing concepts” and “actually building” an agent, or you can call APIs but want to understand the underlying reasoning and decision-making, Agent Craft is designed for you.

Highlights

• Progressive learning path: 15 modules from prompt basics to deployment, each independent and runnable.
• Full-stack coverage: LLM calls, Function Calling, LangChain, RAG, LangGraph, MCP, Swarm/Agents SDK, and Streamlit.
• Runnable code: every module includes commented, locally reproducible code.
• Concepts explained: accompanying CSDN blogs clarify design rationale and mechanisms.
• Real-world practicals: build a customer-service multi-agent cluster and a chat dashboard.