Overview

What is Optimization?

Optimization points a run at a prompt column, scores rewrites against the evals you pick to define what “good” means, and keeps the winner, using the optimizer algorithm you choose to generate and score those rewrites. Reach for it when a prompt is scoring badly and manual tweaking isn’t converging.

Any of six optimizer algorithms can run against the same prompt column and eval set. They’re alternative choices you swap on the same run, each searching for the new prompt differently, which is why choosing an optimizer is worth doing deliberately.

The same engine also runs from inside Simulation, where it optimizes an agent’s prompt using results from a simulation test run instead of a dataset column. See optimization in Simulation for that path.

Keep exploring