datarobot/syftr

About

syftr is an agent optimizer that helps you find the best agentic workflows for a given budget. You bring your own dataset, compose the search space from models and components, and syftr finds the best combination of parameters for your budget. It uses advances in multi-objective Bayesian Optimization and a novel domain-specific "Pareto Pruner" to efficiently sample a search space of agentic and non-agentic flows to estimate a Pareto-frontier (optimal trade-off curve) between accuracy and objectives that compete like cost, latency, throughput.

Please read more details in our blogpost and full technical paper.

We are excited for what you will discover using syftr!

syftr builds on a number…

Excerpted from github.com/datarobot/syftr