SkyDiscover

A Flexible Framework for AI-Driven Scientific and Algorithmic Discovery

SkyDiscover is a modular framework for AI-driven scientific and algorithmic discovery, providing a unified interface for running and comparing algorithms across 200+ optimization tasks. It follows the same high-level evolutionary loop as prior systems like AlphaEvolve, but exposes each stage as a reusable, extensible component:

1. Context Builder. Assembles prompts from the problem spec, prior solutions, and human feedback.
2. Solution Generator. Produces candidates via LLM calls, with optional tool use (e.g., reading codebases).
3. Evaluator. Scores candidates and logs metadata back into the solution database.
4. Solution Selector. Maintains the solution database and picks priors for the next iteration (Top-K, MAP-Elites, adaptive evolutionary methods, etc.).

Come and build easily with SkyDiscover!

🚧 This project is under active development. APIs and interfaces may change.

🏆 Benchmarks & Performance

We implement two adaptive evolutionary algorithms (AdaEvolve and EvoX), achieving:

• Best open-source performance — ~34% median score improvement on 172 Frontier-CS problems over OpenEvolve, GEPA, and ShinkaEvolve
• Matching or exceeding AlphaEvolve and human SOTA — on 8 math and 6 systems optimization tasks
• Real-world discoveries — 41% lower cross-cloud transfer cost, 14% better GPU load balance for MoE serving, 29% lower KV-cache pressure via GPU model placement

🚀 Quick Start

Prerequisites: Python >= 3.10, uv

# Install
uv sync
export OPENAI_API_KEY="<your-key>"

# Try the circle packing benchmark
uv sync --extra math
uv run skydiscover-run benchmarks/math/circle_packing/initial_program.py \
  benchmarks/math/circle_packing/evaluator.py \
  --config benchmarks/math/circle_packing/config.yaml \
  --search adaevolve \
  --iterations 100

# Or run on your own problem
uv run skydiscover-run initial_program.py evaluator.py \
  --search evox \
  --model gpt-5 \
  --iterations 100

Or use the Python API:

from skydiscover import run_discovery

result = run_discovery(
    initial_program="initial_program.py",
    evaluator="evaluator.py",
    search="evox",   # or "adaevolve"
    model="gpt-5",
    iterations=100,
)
print(result.best_score, result.best_solution)

✏️ The Two Files You Write

Scoring Function

Your evaluator is a Python file with an evaluate(program_path) function. It returns a dict with metrics and optional artifacts:

def evaluate(program_path):
    score = run_and_grade(program_path)
    return {
        "combined_score": score,       # primary optimization target (maximized)
        "artifacts": {                 # optional — stored with the solution for future context
            "feedback": "Off by one in the loop boundary",
        },
    }

• combined_score drives evolution. If omitted, SkyDiscover averages all numeric values in the dict.
• artifacts is optional — entries are injected into the next LLM prompt as context.

Starting Solution

Your initial program marks the region to mutate with EVOLVE-BLOCK markers. Everything outside is left untouched.

# EVOLVE-BLOCK-START
def solve(input_data):
    return input_data  # baseline — SkyDiscover will improve this
# EVOLVE-BLOCK-END

If no markers are present, the entire file is treated as mutatable.

🧬 Pick an Algorithm

git clone --depth 1 https://github.com/SakanaAI/ShinkaEvolve.git external_repos/ShinkaEvolve
uv pip install -e external_repos/ShinkaEvolve

⚙️ Configuration

Pass a YAML config with -c. See configs/ for full annotated templates.

max_iterations: 100
llm:
  models: [{ name: "gemini/gemini-3-pro-preview", weight: 1.0 }]
search:
  type: "adaevolve"                  # or "evox", "topk", "beam_search", "best_of_n"
prompt:
  system_message: |
    You are an expert at optimizing algorithms.

API keys (OPENAI_API_KEY, GEMINI_API_KEY, etc.) are resolved from environment variables automatically.

📊 Live Monitor & Human Feedback

Add monitor: { enabled: true } to your config. The dashboard URL prints at run start — scatter plot of all programs, code diffs, metrics, and AI summaries. A Human Feedback panel lets you steer evolution in real time. Replay a completed run:

uv run skydiscover-viewer /path/to/checkpoints/checkpoint_100

📖 Reference

uv run skydiscover-run INITIAL_PROGRAM EVALUATOR [options]

from skydiscover import discover_solution

result = discover_solution(
    initial_solution="def solve(x): return x",
    evaluator=lambda path: {"combined_score": run_tests(path)},
    iterations=50,
    search="evox",
)

--model gpt-5                                               # OpenAI (default)
--model gemini/gemini-3-pro-preview                          # Gemini
--model anthropic/claude-sonnet-4-20250514                   # Anthropic
--model ollama/llama3 --api-base http://localhost:11434/v1   # Local (Ollama, vLLM, etc.)

llm:
  models:
    - name: "gpt-5-mini"
      weight: 0.7
    - name: "gemini/gemini-2.0-flash"
      weight: 0.3

uv sync                              # Base install
uv sync --extra math                 # Math benchmarks (SciPy, JAX, PyWavelets, …)
uv sync --extra adrs                 # ADRS systems benchmarks
uv sync --extra frontier-cs          # Frontier-CS benchmark tooling
uv sync --extra external             # OpenEvolve / GEPA / ShinkaEvolve backends (uv only, not available via pip)
uv sync --extra prompt-optimization  # HotPotQA prompt optimization

🔗 Related Work

SkyDiscover is inspired by AlphaEvolve, Google DeepMind's evolutionary coding agent, and OpenEvolve, an open-source reimplementation of the AlphaEvolve framework.

✍️ Citation

@misc{skydiscover2025,
  title={SkyDiscover: A Flexible Framework for AI-Driven Scientific and Algorithmic Discovery},
  author={Liu, Shu and Cemri, Mert and Agarwal, Shubham and Krentsel, Alexander and Naren, Ashwin and Mang, Qiuyang and Li, Zhifei and Gupta, Akshat and Maheswaran, Monishwaran and Cheng, Audrey and Pan, Melissa and Boneh, Ethan and Ramchandran, Kannan and Sen, Koushik and Dimakis, Alexandros G. and Zaharia, Matei and Stoica, Ion},
  year={2025},
}