Guides

The guides move from composing one typed call to operating larger agent programs safely.

Guide	Main question
Compose agents with Python	How do several agent calls become a workflow?
Repository review workflow	What does a complete typed workflow look like?
Define typed results	How does model output become trusted Python data?
Supply context and instructions	What information changes one execution?
Grant tools and resources	How does an agent interact with external state?
Work with files and commands	How are file changes and processes constrained?
Bound and cancel execution	How does the caller stop further work?
Inspect executions and failures	What evidence survives success or failure?
Test agent programs	How do tests avoid nondeterministic provider calls?