OpenEvolve
NormalDescription
The open-source implementation of AlphaEvolve - an LLM-driven evolutionary coding agent that autonomously discovers new algorithms and optimizes existing code.
Key Features
- Evolutionary code search - LLM-driven island model that evolves code variants in parallel
- Multi-objective Pareto optimization - jointly optimizes speed, memory, and correctness
- Real-hardware speedups - 2-3x faster GPU kernels, matrix multiplications, and circle packings
- Reproducible experiments - full evaluation pipelines with research-grade metrics
- Model agnostic - works with OpenAI, Anthropic, Gemini, and any LLM backend
- HTML visualization - inspect the full solution tree and per-step code in a browser
Use Cases
💡 Auto-optimize performance-critical code - GPU kernels, serializers, math libraries
💡 Search for top-scoring solutions on Kaggle / MLE-bench-style ML competitions
💡 Discover new algorithmic approaches to heuristics, approximation, and combinatorial problems
💡 Use as a reproducible baseline for AlphaEvolve-style research
💡 Wire LLMs into CI for an automated code-optimization pipeline
Strengths & Limitations
✅ Strengths
- • High community interest (7.3k stars)
- • Permissive open-source license (Apache-2.0)
Categories
Quick Start
pip install -U openevolve
export OPENAI_API_KEY=<your-key>
# Run a built-in example: circle packing
python -m openevolve examples/circle_packing/initial_program.py \
examples/circle_packing/evaluator.py \
--iterations 100
# Open logs/<run>/tree.html to inspect the evolution tree