CodeEvolve: Open-Source Evolutionary Coding Agent for Algorithmic Discovery
An open-source evolutionary coding agent that combines LLMs, island-based search, crossover, meta-prompting, refinement, and evaluator feedback for algorithmic discovery.
How does island-based evolutionary search compare with simpler optimization methods when evaluator calls and model inference are held to the same compute budget?
CodeEvolve is important because it is a public implementation of the AlphaEvolve-style thesis: when a problem can be scored automatically, a coding agent can search the program space and retain improvements.
Technical shape
The framework uses an island-based evolutionary search over candidate programs, LLM-generated mutations, crossover, meta-prompting, refinement, and a structured archive. The paper reports competitive results on AlphaEvolve-style benchmark tasks under matched conditions.
Build note
This is a practical starting point for ASI-adjacent experimentation because it replaces vague “AI creativity” with a measurable loop: propose code, execute tests, score the output, and keep only improvements.