Each tutorial builds one agent, step by step, with a real model. You write the files of each tutorial in its own folder in my_tutorials/, a directory that git ignores. The complete code of each tutorial is in the same folder in examples/tutorials, and CI runs it.

Tutorial	You build	You learn
1. Your first agent	A town assistant that answers questions from the town documents	`ReActAgent`, `ReActConfig`, `@tool`, threads and the trace
2. Cite the documents	The same assistant, with a citation for each fact	Elements, `bind`, `Recorded`, element numbers, limits, citations, tool tests
3. One agent for many users	One assistant that searches only the documents that each user can read	Resources, a config for each turn, prompt files, the agent and the graph
4. Control what the model sees	The same assistant, with a context policy for long conversations	The context builder, what each model call receives, the context policy, `reopen`
5. A refund agent with approval and a safe payment	A refund desk that asks an employee for approval and pays once	Agents and graphs of your own, routers, requests, effect steps and recovery
6. Serve an agent and chat with it	A server for the agent of tutorial 3, and a command-line chat client	The thread API, resources from the HTTP request, server-sent events, streamed answers

Do the tutorials in order. Tutorials 1 to 4 build one assistant, one step at a time. Tutorial 5 builds a new agent. Tutorial 6 serves the agent of tutorial 3.

You need Python 3.12 or later, uv, and a chat model with tool calls behind an OpenAI-compatible API.