CookbookGet started

Cookbook

Short, task-oriented recipes — the fastest path from zero to a working result. Each one is copy-paste first, with just enough explanation to adapt it to your own agent. Start with GEPA if you want to make an agent measurably better.

Optimize

Optimize a prompt with GEPATake a weak prompt, give GEPA a dataset and a verifier, and ship a measurably better one.Score outputs with a verifierTurn pass / fail rules into a verifier that scores every output the same way.

Observe

Trace an agent end to endInstrument an agent in two calls so every step, tool call, and token is captured.

Evaluate

Run a benchmark suitePull a shared task package and run it against your agent endpoint.