openevolve

Open-source implementation of AlphaEvolve

OpenAlpha_Evolve is an open-source Python framework inspired by the groundbreaking research on autonomous coding agents like DeepMind's AlphaEvolve.

A next-generation automatic algorithm design platform, making automated algorithm design more accessible and easier to use

[ICML'26] Towards Feedback-to-Plan Decisions for Self-Evolving LLM Agents in CUDA Kernel Generation

AI4Math skills for search-based experiments, OpenEvolve workflows, and iterative program improvement.

Executable benchmark for AI-driven potential discovery for the k-server conjecture, with evaluators, metric instances, and released experiment workflows.

Correctness oracle for AI code-optimization loops: catches reward hacks with canonical, metamorphic, withheld-input, and timing checks.

Codex Skill and platform-neutral research CLI for reproducible, auditable OpenEvolve algorithm discovery with holdouts, ablations, budgets, and novelty audits.

Evaluation-first AI case study for evolving retry/backoff policies with local LLMs, strict QA gates, and holdout validation.

Using Levi to optimize randomize algorithms for set membership and tasks in similar vein

Reproducing AlphaEvolve's SOTA bound on the Erdős minimum-overlap problem with an evolutionary LLM pipeline (OpenEvolve) — converges in ~10 iterations, two independent backends, one-command verifier

NoemaEvolve extends LLM evolutionary program search with RL based reflection and a ReAct style coordination

A general optimisation/search substrate: a pure hexagonal core + pluggable search strategies + pluggable compute backends.

A production fork of OpenEvolve: the engine byte-identical to upstream, plus a real-time control plane, a provider broker with role-based routing, and an isolated agent sandbox.