Start here

The curriculum runs in rough dependency order, so reading top to bottom works. If you would rather go straight at a problem, pick the row that matches.

If you have never built an agent

Read What an agent actually is, then build the twenty-line loop in it by hand before touching a framework. Then read The model is stateless. Those two pages together are the foundation everything else assumes.

If your agent works but you do not know why

Start with The model is stateless, then Anatomy of the agent loop. Most of the mystery is in what your code is re-sending on each turn.

If the model keeps picking the wrong tool

Go straight to Tool calling: schemas and structured output. The fix is almost always in the tool description, not the system prompt.

If it worked in the playground but not in your app

The model is stateless. The two are sending different context. They always are.

How to read this

Pages are short on purpose and assume you will go deeper with a model. Each one ends with the specific prompts worth pasting, aimed at your stack rather than a generic example.