Skip to main content
Automatically improve your recipes using AI judge feedback.
The optimizer runs your recipe, evaluates the output, and applies improvements iteratively.

Quick Start

How It Works

1

Run

Executes the recipe and captures output
2

Judge

AI evaluates task achievement, output quality, and instruction following
3

Improve

Proposes specific YAML changes based on feedback
4

Apply

Updates agents.yaml with improvements
5

Repeat

Continues until score threshold or max iterations

Options

Examples

Optimize with defaults:

What Gets Improved

The AI judge evaluates and improves:
Did the agent accomplish what it was asked to do?
Does the output match expected format and contain useful information?
Did the agent follow specific instructions, format, and constraints?
How well did the agent handle errors and edge cases?

Score Interpretation

Start with default threshold (8.0) and increase to 9.0 for production-critical recipes.

Workflow

Next Steps

Recipe Judge

Manually judge recipe traces

Recipe Registry

Share optimized recipes