llm-router

What This Skill Does

The OpenClaw LLM Router skill serves as a high-performance, unified interface for autonomous agents, acting as a gateway to over 70 state-of-the-art AI models. By providing a single OpenAI-compatible API endpoint, it abstracts the complexity of managing disparate API keys, billing accounts, and SDKs. Developers can seamlessly integrate top-tier intelligence from OpenAI (GPT series), Anthropic (Claude), Google (Gemini), Alibaba (Qwen), DeepSeek, and xAI (Grok) into their agents without writing provider-specific boilerplate code. The router enables advanced features like automatic fallback strategies, model-based routing for cost optimization, and simplified multi-model comparisons, ensuring that your agent always has the right tool for the task at hand.

Installation

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

clawhub install openclaw/skills/skills/renning22/asia-llm-router-skills

Once installed, you must configure your environment by setting the AISA_API_KEY. You can obtain this key from the AIsa platform dashboard. Set it using: export AISA_API_KEY="your-key". This setup allows your agents to interface directly with the centralized router API at https://api.aisa.one/v1/chat/completions.

Use Cases

• Cost-Efficient Scaling: Route trivial tasks (e.g., summarization, formatting) to low-cost, high-speed models, and reserve computationally expensive models like GPT-4 or DeepSeek-R1 for complex reasoning and logic tasks.
• Robust Reliability: Implement automated fallback logic where an agent attempts a request on a premium model and, upon failure or rate-limiting, immediately rotates to a secondary or tertiary model provider.
• Comparative Analysis: Run parallel evaluations where the agent prompts three different models with the same input to reach a consensus or select the most accurate response for a critical application.
• Model Agnosticism: Build your agent architecture once using standard OpenAI schemas and switch underlying model providers via a simple parameter change rather than refactoring your core codebase.

Example Prompts

1. "Use GPT-4 to draft a project proposal, then ask Claude to rewrite it in a more persuasive tone."
2. "Compare the code quality of DeepSeek-V3 and Qwen-2.5 for this specific Python function: [insert code block]."
3. "If the primary model fails to generate a response, please automatically fallback to Gemini-1.5-Pro and notify me."

Tips & Limitations

• Tip: Monitor your latency and tokens closely via the AIsa dashboard to optimize your model selection for specific workflows.
• Tip: Use the 'system' role strategically to define the persona regardless of which model you are routing to.
• Limitation: Ensure you check the official AIsa marketplace documentation for specific model availability and tier restrictions, as some high-compute models may be subject to stricter rate limits.