fugusashi-router

Fugusashi Router

<div align="center">

By eulogik — building AI infrastructure for everyone.

</div>

---

What is This?

CMA-ES evolved routing weights for the Fugusashi intelligent model router. A 385-dimensional weight vector that maps prompt embeddings to model selection — learned via Covariance Matrix Adaptation Evolution Strategy, the same approach used in Sakana AI's TRINITY paper.

Like Sakana Fugu. But Free. And Yours.

Usage

import numpy as np
import json
from sentence_transformers import SentenceTransformer

# Load weights
with open("cmaes_weights.json") as f:
    data = json.load(f)

weights = np.array(data["mean"])
bias = data["mean"][-1]

# Embed prompt and route
model = SentenceTransformer("all-MiniLM-L6-v2")
embedding = model.encode(["Write a Python function to sort a list"], normalize_embeddings=True)[0]

logits = embedding * weights[:384] + bias
models = ["gpt-oss-120b", "nemotron-3-ultra", "nemotron-3-super", "hermes-3-405b", "lfm-2.5-1.2b"]
chosen = models[np.argmax(logits)]
print(f"Route to: {chosen}")

Training Details

Project Links

Citation

If you use Fugusashi in your work:

@software{fugusashi2026,
  title={Fugusashi: Open-Source Intelligent Model Router},
  author={{eulogik}},
  year={2026},
  url={https://github.com/eulogik/fugusashi}
}

License

MIT — use it however you want.

---

<div align="center">**Built with ❤️ by [eulogik](https://eulogik.com)**</div>