A free, open, research-backed curriculum for people who want to actually build with AI agents — not just chat with them. Every tutorial has runnable code, copy-paste setup, and honest tradeoffs.
Why this exists: the agentic-AI space is loud and full of hype. This repo is the opposite — practical, tested, and grounded in what the ecosystem actually uses (Ollama, LangChain/LangGraph, AutoGen, MCP, LlamaIndex). Stars on those projects are the demand signal; the tutorials follow it.
| # | Tutorial | Teaches | Demand signal |
|---|---|---|---|
| 00 | Start Here | How to use this repo + pick your path | — |
| 01 | Run AI Locally with Ollama | Private, free inference on your machine | Ollama · 176k★ |
| 02 | Build Your First Agent (from scratch) | ReAct loop, tools, no frameworks | Foundations |
| 03 | Production Agents with LangGraph | Stateful graphs, memory, human-in-loop | LangChain 141k★ · LangGraph 37k★ |
| 04 | Multi-Agent Systems with AutoGen | Agent teams that solve tasks together | AutoGen 60k★ |
| 05 | MCP: Build a Tool Server | The new standard for agent tool-use | MCP 88k★ |
| 06 | RAG with LlamaIndex | Give agents knowledge from your docs | LlamaIndex 51k★ |
| 07 | Prompt Engineering That Works | The skill under everything | — |
| 08 | Ship Agents to Production | Eval, guardrails, cost, observability | — |
| 09 | Agent Skills | Reusable capability packages | Anthropic skills 160k★ |
Start at 00-start-here.md. If you've never run a model, do 01 → 02. If you want production patterns, jump to 03/04/05. Every folder is self-contained.
- Run it, don't just read it. Every code block is tested.
- Show the tradeoffs. Frameworks are not always the answer (02 builds one with zero deps).
- Local-first. Privacy and cost matter; 01 gets you free local inference.
See CONTRIBUTING.md. Corrections and new tutorials welcome — especially ones with runnable code.
MIT. Use it, fork it, teach with it.