model-router

What This Skill Does

The model-router skill is an intelligent orchestration layer designed to optimize OpenClaw's computational expenditure. Instead of defaulting to the most expensive or powerful LLM for every interaction, model-router acts as a traffic controller. By analyzing the inherent complexity and objective requirements of a task, it dynamically selects the most cost-effective model from your configured providers. Whether you are performing simple data transformations or debugging complex system architectures, this skill ensures that you are using the minimum required intelligence, thereby drastically reducing latency and costs without sacrificing output quality.

Installation

To integrate the model-router into your agent's workflow, execute the following command in your terminal: clawhub install openclaw/skills/skills/globalcaos/smart-model-router

Once installed, ensure your agent environment is configured with the necessary API keys for your preferred tiers (flash, fast, mid, strong, reasoning).

Use Cases

• Autonomous Agent Spawning: When an agent needs to spin up sub-agents to handle specific sub-tasks, model-router determines the model tier for those workers.
• Cron Job Optimization: Automate routine system reports or data synchronization tasks by selecting 'flash' or 'fast' models, saving resources for heavy-duty analytical tasks.
• Ambiguous Request Handling: Use the built-in classifier for natural language prompts that don't clearly fall into a single category, ensuring that a 'strong' or 'reasoning' model is only invoked when truly necessary.

Example Prompts

1. "Route this task: 'Convert this list of 500 email addresses into a clean JSON array' using the most efficient model."
2. "I need to debug this failing authentication module across three different files; select the best model for this deep investigation."
3. "Summarize the following document into bullet points for a quick update, keeping costs as low as possible."

Tips & Limitations

• The 'Good Enough' Principle: Always prefer the 'flash' or 'fast' tiers for mechanical tasks like string manipulation or date lookups. Over-engineering by selecting an 'opus' or 'o3' model for simple tasks is a common performance bottleneck.
• Classification Latency: While the router itself is fast, using the classifier adds a minor overhead. For known routine tasks, bypass the classifier by specifying the tier manually.
• Model Availability: Ensure your provider supports the models mapped in the Tier table; if a specific model is unreachable, the router will fallback to the next available tier.