DataRobot announced syftr in May 2025, an open-source framework that automatically searches through millions of possible AI workflow configurations to identify the combination of components—such as retrievers, language models, and prompt strategies—that best balances accuracy, speed, and cost for a given enterprise use case. The framework is available publicly; an enterprise version was planned for fall 2025.

Syftr uses multi-objective Bayesian Optimization to efficiently search a configuration space of over 10²³ possible agentic workflow combinations without requiring manual testing of each option. It is component-agnostic, meaning it can evaluate any embedding model, LLM, or pipeline module. In testing against six industry RAG benchmarks, the framework increased accuracy by 25 percent overall while reducing costs by 37 percent at baseline accuracy levels. Syftr generates production-ready LlamaIndex pipeline code for the optimal configuration identified. The open-source framework was released on Hugging Face; a DataRobot enterprise-integrated version was announced for fall 2025.