holny

Evolve Router — AI skill for Claude Code

AI community

A local-first LLM routing gateway that thinks before it routes — a decision model (Jev / laya) picks the most suitable model for every request.

How to install Evolve Router

This entry records only its repository, not the path inside it, so there is no exact command to give. Open holny/evolve-router and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Evolve Router does

A local-first LLM routing gateway that thinks before it routes — a decision model (Jev / laya) picks the most suitable model for every request. Self-evolving flywheel learns from every outcome. Scientific formula, fully auditable. One 6.7 MB Rust binary, zero cloud. Supports opencode, Claude Code, Codex, Pi, openclaw, hermes — OpenAI & Anthropic.

Alternatives in AI

Khazix Skills — DA collection of AI Skills for managing and evolving your skill library 1.5k ★
Orkas — Open-source multi-agent AI desktop client — build and command your AI agent team through conversation 1.5k ★
Phantom — An AI co-worker with its own computer 1.3k ★

README

EvolveRouter

**A local LLM routing gateway with a decision model at its core. It evolves with every request.**

[![License](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE) [![Rust](https://img.shields.io/badge/rust-stable-orange.svg)](https://www.rust-lang.org) [![Tests](https://img.shields.io/badge/tests-111%20passing-brightgreen.svg)](#development)

[English](README.md) | [中文](README.zh-CN.md)

**Not the best model — the right one, for every request.**

A local-first LLM routing gateway: a decision model judges every request's task profile, a flywheel learns from every outcome, and a scientific formula keeps every decision traceable. **Your data never leaves your machine.**

![EvolveRouter architecture](docs/assets/architecture.svg)

**EvolveRouter** is a local-first routing gateway that sits between your Agent and every LLM provider you use. It doesn't ship every request to the "best" model — it routes each one to the **most suitable** one, judged on the fly by a **decision model** (any System-One-compatible backend — cloud or self-hosted — plus a multilingual heuristic second opinion, take-conservative fusion) against what the task actually needs and what your providers can afford right now. Every request feeds the flywheel, so the routing gets sharper the more you use it.

What makes it different: every decision is **local, transparent, and auditable** — JSONL event log, response headers (`x-ev-model / x-ev-reason / x-ev-decision-id`), zero-build dashboard. User data never leaves the host. And when the decision model is unavailable, the heuristic takes over and **routing never blocks**.

Why not another proxy

	EvolveRouter	LiteLLM / OpenRouter	RouteLLM
Routing decision	Decision model judges every request (8 signals) + heuristic second opinion, fail-open	Config/prefix rules, or model-based router trained offline	Trained router models, offline
**Self-evolutio