>_
prompt-opt

Prompt
optimization.

One shared engine, many problems, many algorithms_

see gepa >

synth docs / prompt optimization

Wrap your application in a standardized container route contract.

Straightforwardly evaluate and run against all optimizers. Any language, any software, any model.

read containers guide >

01 / guide

Containers

>

Task containers expose program, rollout, checkpoint, resume, verifier, and task-info surfaces for optimizers.

> /program
> /rollout
> /checkpoint
> /resume

optimization algorithms / public cookbooks

Pick the right optimizer for the job.

Choose the algorithm that best fits your software and priorities. Easily swap out algorithms and reconfigure runs.

read docs >
algorithm
Algorithm: GEPA / Genetic Evolutionary Prompt Optimization
public cookbooks
banking77classificationview >
minigrid*embodied agentview >
tblite*coding agentview >
criteria
Agent policy first
Codex proposer
Flash evolve throughput
see gepa >

Our research notes.

Findings and devlog.

read the blog >