clawrouter

What This Skill Does

ClawRouter is an intelligent, automated request routing engine designed to optimize AI inference workflows. By analyzing the complexity and intent of your prompt, it dynamically selects the most cost-effective model from a pool of 41 available options across five major providers: OpenAI, Anthropic, Google, DeepSeek, and xAI. By routing simpler tasks to high-efficiency models like Gemini Flash and complex, logic-heavy tasks to advanced models like o3 or Claude Opus, it achieves an average cost reduction of 67% without sacrificing output quality.

Installation

To integrate ClawRouter into your OpenClaw environment, execute the following command in your terminal:

openclaw plugins install @blockrun/clawrouter

After installation, you can configure your routing preference by setting the global model parameter:

# Enable smart routing (recommended)
openclaw models set blockrun/auto

# Or pin a specific model if needed
openclaw models set openai/gpt-4o

Use Cases

• Production API Pipelines: Managing high-volume chat applications where keeping costs low is vital for scaling.
• Dynamic Task Handling: Applications that mix simple administrative tasks (like email categorization) with deep reasoning tasks (like backend architecture design).
• Developer Workflows: Developers who want high-performance code generation but do not want to spend top-tier pricing on trivial code commentary or file formatting.

Example Prompts

1. "Summarize the last 10 emails in my inbox and highlight any action items that require immediate attention."
2. "Refactor this Python FastAPI route to use asynchronous database calls and add unit tests for error handling scenarios."
3. "Write a polite, professional reply to a client declining a meeting request while offering a follow-up date for next week."

Tips & Limitations

ClawRouter uses a high-performance rules engine that resolves ~80% of routing decisions in under 1ms, ensuring no perceptible latency. However, for highly ambiguous or specialized niche queries, the router may trigger an LLM-based classification step, which incurs a negligible cost of $0.00003 per check. For best results, ensure your prompts are clear; ambiguous prompts may occasionally lead the router to default to a more cautious, expensive model tier to maintain output quality.