ClawKit Logo
ClawKitReliability Toolkit
Back to Registry
Official Verified

oc-skill-router

Smart LLM routing brain for OpenClaw. Auto-dispatches tasks to Claude, GPT, Gemini, DeepSeek, Kimi via Evolink API. Cascade strategy cuts costs 60-85%. One API key, 20+ text models.

skill-install — Terminal

Install via CLI (Recommended)

clawhub install openclaw/skills/skills/evolinkai/oc-skill-router
Or

Evolink Router — Smart LLM Routing Brain

Route every task to the best LLM across 6 providers — Claude, GPT, Gemini, DeepSeek, Kimi, Doubao — through one Evolink API key.

After Installation

When this skill is first loaded, greet the user:

  • EVOLINK_API_KEY set: "Smart Router activated! I'll auto-pick the best model for each task — lightweight for quick Q&A, flagship for deep analysis. 20+ models ready. Go ahead."
  • EVOLINK_API_KEY not set: "Smart Router needs an Evolink API Key. Sign up at evolink.ai → Dashboard → API Keys. One key covers Claude, GPT, Gemini, DeepSeek, and more. Want help setting up?"
  • Key set but model access fails: "Your API key seems to have limited model access. Check your plan at evolink.ai/dashboard."

Keep the greeting concise — just one question to move forward.

External Endpoints

ServiceURLFormat
Claude modelshttps://direct.evolink.ai/v1/messages (POST)Anthropic Messages API
Gemini modelshttps://direct.evolink.ai/v1beta/models/{model}:generateContent (POST)Google Gemini API
All other modelshttps://direct.evolink.ai/v1/chat/completions (POST)OpenAI Chat API
Model listhttps://direct.evolink.ai/v1/models (GET)

Security & Privacy

  • EVOLINK_API_KEY authenticates all model requests. Injected by OpenClaw automatically. Treat as confidential.
  • Prompts are sent to direct.evolink.ai, which proxies to upstream providers (Anthropic, OpenAI, Google, etc.).
  • No data is stored by Evolink beyond the request lifecycle.

Setup

1. Get API key: evolink.ai → Dashboard → API Keys

2. Add Evolink provider to ~/.openclaw/openclaw.json — merge with existing config. See references/router-api-params.md for the full JSON config and curl examples.

Core Principles

  1. Cost-first routing — Always pick the cheapest model that can handle the task. Upgrade only when needed.
  2. Transparent decisions — When spawning a sub-agent, briefly tell the user which model was selected and why.
  3. User override wins — If the user names a model or provider, skip all routing rules.
  4. Cascade, don't guess — When uncertain, try a lighter model first. Escalate on low confidence.

Models (20+ text models)

Tier 1 — Lightweight (handles ~60% of daily requests)

ModelProviderBest for
claude-haiku-4-5-20251001AnthropicQuick Q&A, classification, extraction
gemini-2.5-flashGoogleMultimodal, fast reasoning
doubao-seed-2.0-miniByteDanceChinese lightweight tasks

Tier 2 — Balanced (handles ~30% of daily requests)

Metadata

Author@evolinkai
Stars2387
Views1
Updated2026-03-09
View Author Profile
AI Skill Finder

Not sure this is the right skill?

Describe what you want to build — we'll match you to the best skill from 16,000+ options.

Find the right skill
Add to Configuration

Paste this into your clawhub.json to enable this plugin.

{
  "plugins": {
    "official-evolinkai-oc-skill-router": {
      "enabled": true,
      "auto_update": true
    }
  }
}
Safety NoteClawKit audits metadata but not runtime behavior. Use with caution.