optimize-anything

To iteratively improve a system prompt, define the candidate prompt, an evaluator (Python, shell, or LLM judge), and an objective. The optimization process refines the text artifact through cycles of evaluation and LLM-driven improvement.

Yes, you can optimize code and configurations by defining the text artifact as a candidate and providing an evaluation script. The system uses LLM-powered search and evaluation feedback to iteratively refine the configuration or code.

Setting up an evaluator for text optimization requires defining an objective and specifying an evaluator, which can be a Python script, a shell command, or an LLM judge. You also need a candidate artifact to evaluate.

Prompt engineering optimization supports single-task, multi-task, and generalization optimization modes. These modes allow you to refine text artifacts for specific tasks or broader applicability using iterative LLM-powered search.

The best way to automate agent architecture refinement is treating the architecture as a text artifact, defining an objective, and using an evaluator. The iterative LLM-driven search evaluates and modifies the architecture to improve performance.

Iterative text optimization requires a defined objective because it serves as the target for the evaluator (Python, shell, or LLM judge). The objective guides the LLM-powered search to effectively refine the candidate artifact through feedback.