This repository implements the Darwin Gödel Machine, a system enabling open-ended evolution of self-improving agents. The system uses large language models (OpenAI, Anthropic) to generate code modifications, then empirically validates each change against coding benchmarks like SWE-bench. Agents can improve their own codebase, including the code responsible for making further improvements, creating a recursive self-improvement loop. Docker is used to safely sandbox code execution during validation.