AlgoDisco: Method Implementations and Tools for
LLM-driven Automated Algorithm Design

✨ Key features

• Multiple Search Methods: FunSearch, OpenEvolve, EoH, (1+1)-EPS, RandSample
• Multiple LLM Providers: OpenAI, Claude, vLLM, SGLang
• Sandboxed Execution: Safe evaluation of generated algorithms
• Flexible Configuration: YAML-based configuration system

🛠️ Requirements

• Python >= 3.11 (recommended: 3.12)

🚀 Quick start

# Install the package in editable mode
pip install -e .

If you want to use the default OpenAI path, export your API key first:

export OPENAI_API_KEY="your-api-key"

We provide two practical starting points:

1. Python API example:

   python examples/online_bin_packing/run_funsearch.py
   

2. YAML example (recommended for learning the config system):

   bash examples/run_online_bin_packing.sh funsearch
   

For more details, see Quick Start.

📖 Documentation

• Installation
• Search Methods
• LLM Providers
• API Reference
• Tutorial: Create Your Own Problem

🔍 Available search methods

• FunSearch
• OpenEvolve
• EoH
• (1+1)-EPS
• RandSample

💡 Quick example

Want to run a search? Here's how:

Option 1: Use provided example (Online Bin Packing)

We provide a ready-to-use example in examples/online_bin_packing/.

Python-style (direct code):

python examples/online_bin_packing/run_funsearch.py

Set your API key first:

export OPENAI_API_KEY="your-api-key"

YAML-style (recommended):

Configs for each method are in examples/online_bin_packing/configs/:

1. Copy and edit the config:

   cp examples/online_bin_packing/configs/funsearch.yaml examples/online_bin_packing/configs/my_config.yaml
   

2. Open my_config.yaml and replace:

   • api_key: null → api_key: "your-openai-key" (or set OPENAI_API_KEY env var)

3. Run with any method (just change the argument):

   # Run with FunSearch
   bash examples/run_online_bin_packing.sh funsearch
   
   # Run with OpenEvolve
   bash examples/run_online_bin_packing.sh openevolve
   
   # Run with EoH
   bash examples/run_online_bin_packing.sh eoh
   
   # Run with (1+1)-EPS
   bash examples/run_online_bin_packing.sh one_plus_one_eps
   
   # Run with RandSample
   bash examples/run_online_bin_packing.sh randsample
   

Available methods: funsearch, openevolve, eoh, one_plus_one_eps, randsample

For SwanLab integration, use funsearch_swanlab:

bash examples/run_online_bin_packing.sh funsearch_swanlab

Option 2: Create your own experiment

1. Choose a method (e.g., FunSearch) and copy one of the example configs:

   cp examples/online_bin_packing/configs/funsearch.yaml my_experiment.yaml
   

2. Edit my_experiment.yaml to set your:

   • template_program_path: Your algorithm template
   • task_description_path: Problem description
   • evaluator.class_path: Your evaluator class
   • LLM provider settings (API key, model, etc.)

3. Run:

   python -m algodisco.methods.funsearch.main_funsearch --config my_experiment.yaml
   

📝 Citation

If you use AlgoDisco in your research, please cite:

@misc{algodisco,
  title = {AlgoDisco: Method Implementations and Tools for LLM-driven Automated Algorithm Design},
  author = {Rui Zhang},
  year = {2026},
  url = {https://github.com/RayZhhh/algodisco},
}

📜 License

MIT License - see LICENSE for details.