darwinian-evolver

To optimize LLM prompts automatically, you can run an evolutionary search loop around a candidate prompt to progressively improve its performance against a defined fitness metric. This requires defining an Organism, Evaluator, and Mutator to drive the optimization.

Yes, you can improve regex patterns and SQL queries by running the evolutionary search loop around those candidate artifacts. The loop progressively mutates and evaluates the regex or SQL to optimize performance against your specific fitness metric.

Before running an evolutionary code improvement loop, you need Python 3.11 or higher, git, uv, and access keys for your chosen LLM provider. These dependencies are required to execute the minimal Organism, Evaluator, and Mutator abstractions.

The best way to define a custom problem for evolutionary search is using the built-in problem-definition scaffolding. You implement the Organism, Evaluator, and Mutator abstractions and drive the loop via the CLI or a custom Python driver.

Yes, the evolutionary search loop works with OpenRouter through a provided parrot evolution driver. You can run this OpenRouter-based driver to start a small run and inspect the optimization results in the output directory.

You should not use an evolutionary search loop for prompt engineering if your optimization problem cannot be clearly defined by a fitness metric or if you lack the required Python 3.11, git, uv, and LLM provider access keys needed to run the loop.