Optimize_anything: A Universal API for Optimizing Any Text Parameter (gepa-ai.github.io)

The launch of the optimize_anything API provides a groundbreaking tool for optimizing any text-based artifacts, such as code, prompts, and configurations. Building on the capabilities of GEPA, this declarative API allows users to specify what they want to optimize and how to measure success, while the system takes care of the search process. In extensive tests, optimize_anything has shown to outperform specialized tools, cutting costs significantly in cloud scheduling by 40% and improving task completion speeds by 47%, highlighting its versatility across various domains. What sets optimize_anything apart is its ability to unify three optimization paradigms—single-task, multi-task, and generalization—under one API, enabling it to tackle complex challenges like discovering new agent architectures or refining prompts for unseen examples. It incorporates a novel concept called Actionable Side Information (ASI), which allows diagnostic feedback to be used effectively in optimization processes, enhancing the feedback loop for an LLM’s reflective reasoning. This system redefines how optimization tasks are formulated, moving beyond traditional methods by allowing a wide range of text artifacts to be improved, indicating a significant advancement for AI/ML applications.