Model Router is a self-aware routing engine for OpenClaw that dynamically matches specific tasks to the most capable and cost-effective AI models available in your configuration.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install sonnet-model-router
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install sonnet-model-router using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Model Router?
Model Router for Openclaw Skills is a sophisticated management layer that transforms static model selections into a dynamic, performance-oriented ecosystem. By maintaining a living model registry, it evaluates the strengths of Claude, Gemini, GPT, and DeepSeek based on real-world benchmarks rather than just leaderboards. It ensures that high-stakes reasoning tasks go to premium models while routine background processes are handled by economy-tier models, drastically reducing API costs without sacrificing quality.
The skill operates with a privacy-first approach, interacting only with model IDs and provider names from your gateway configuration without ever touching sensitive API credentials. It empowers users to switch between Performance, Balanced, and Economy modes through simple natural language commands, making it an essential tool for developers managing multi-provider AI environments.
Model Router Use Cases
- Automatically switching to Claude Opus 4.6 for complex architectural decisions or deep debugging sessions.
- Routing background subagent tasks to Gemini Flash or DeepSeek to minimize operational costs during high-volume processing.
- Receiving proactive recommendations to update routing rules when new, more efficient models are detected in your provider configuration.
- Overriding default models for specific tasks like web browsing or financial analysis based on proven benchmark superiority.
How Model Router Works
- The skill performs a non-invasive scan of the OpenClaw gateway configuration to identify active providers and model IDs.
- It compares the active setup against a local
model-registry.jsonfile which stores benchmark data (SWE-bench, OSWorld, etc.) and pricing. - For new models, it triggers a user-authorized web search to fetch the latest performance data and classification tiers.
- It suggests an optimal routing mode (Performance, Balanced, or Economy) based on the user's current needs and available keys.
- Upon user approval, it applies permanent configuration patches or per-session overrides to ensure the right model handles the right task.
Model Router Setup
To integrate this capability into your environment, ensure you are running the required OpenClaw version and have at least one provider configured.
# Ensure your OpenClaw instance is up to date
docker pull openclaw/openclaw:latest
# Or via Git
openclaw update
# Initialize the model router by asking the agent:
# "Check my available models and setup routing."
Model Router Data Schema & Taxonomy
The skill manages a local model-registry.json file to track benchmarks and routing logic.
| Feature | Description |
|---|---|
| Model Metadata | Tracks provider, pricing per 1M tokens, and context window size. |
| Benchmarks | Stores scores for SWE-bench, OSWorld, GPQA, and ARC-AGI-2. |
| Routing Rules | Maps task types (e.g., computer-use, deep-reasoning) to specific model IDs. |
| Tiers | Classifies models into premium, mid, economy, or free based on cost. |
Model Router Advanced Features
- Adaptive triggers that respond to natural language cues like "work harder" or "save money" to shift routing logic on the fly.
- Automatic task-specific hard routing for domains where specific models have a significant benchmark lead (e.g., Claude for OSWorld tasks).
- Self-learning registry updates that flag benchmark data for refresh every 90 days to stay current with AI industry shifts.
- Multi-agent orchestration support, allowing primary agents and subagents to run on different providers simultaneously.
SKILL.md
Loading
Related Openclaw Skills
kk-daily-backup-report
A mobile-optimized reporting skill that provides concise, categorized daily backup status updates for servers and personal data.
GitHub AI Trends
A specialized tool for generating formatted leaderboards of trending GitHub AI projects based on real-time star counts and activity.
Desktop Control
A comprehensive desktop automation skill that enables AI agents to control the mouse, keyboard, and screen across multiple operating systems.
Feishu Image Sender
A specialized automation skill for sending local image files from your workspace directly to Feishu channels.
Crypto & Stock Market Data (Node.js)
A professional-grade financial data integration for AI agents providing real-time crypto and stock market metrics without the need for external dependencies.
Smart Spawn API
A zero-plugin HTTP API that dynamically selects the most cost-effective and high-performing AI models for any specific task or sub-agent spawn.
Featured*
