model-router

What This Skill Does

The Model Router is a sophisticated orchestration layer for OpenClaw that empowers your AI agent to make intelligent, cost-aware decisions about which LLM to deploy for any given task. Rather than relying on a static configuration, the Model Router maintains a living, self-updating registry of your available providers—including Claude, Gemini, GPT, and DeepSeek. It evaluates the complexity of your request against real-world benchmark data and cost metrics stored in the 'model-registry.json' file, ensuring that high-intensity reasoning tasks receive premium compute while mundane summarization tasks are routed to the most cost-effective model available.

Installation

To integrate this intelligence into your OpenClaw environment, run the following command in your terminal: clawhub install openclaw/skills/skills/chandika/sonnet-model-router

Use Cases

• Dynamic Task Routing: Automatically switch between 'deep-reasoning' models for complex architectural tasks and 'cheapest-available' models for simple draft generation.
• Cost Optimization: Maintain a balance between performance and budget by letting the agent suggest models based on current pricing and context limits.
• Benchmarking-Driven Selection: Leverage empirical data from the registry to ensure your coding tasks are handled by the models with the highest SWE-bench scores currently in your stack.

Example Prompts

1. "Scan my gateway configuration and update the model registry with the latest benchmark data for my new providers."
2. "I need to refactor this entire codebase; route this task to the best available model for deep-reasoning and complex coding."
3. "Summarize these logs for me using the cheapest possible model available in my registry to save on token costs."

Tips & Limitations

• Manual Trigger Only: The router is intentionally designed to be non-intrusive. It never scans your environment on startup or heartbeats to preserve your privacy and performance. You must manually invoke a scan when you add new API providers.
• Privacy First: This skill is strictly local. It never transmits your API keys or credentials; it only reads the metadata of the providers already configured in your gateway.
• Registry Maintenance: While the skill handles the creation of 'model-registry.json', advanced users can manually edit the routing rules within this file to force specific models for custom workflows.