edge-router

What This Skill Does

The edge-router skill acts as an intelligent orchestration layer for OpenClaw AI agents, enabling them to dynamically distribute computational workloads across a hybrid infrastructure. By leveraging a tiered priority system, the router automatically evaluates the cost and availability of compute backends, ranging from local Ollama instances and Vast.ai cloud GPUs to Wukong 72Q quantum hardware. This abstraction allows developers to define task requirements rather than hardcoding infrastructure dependencies, ensuring optimal cost-efficiency and performance for tasks like model inference, model training, or specialized quantum computing workloads. It effectively manages the lifecycle of a task from routing recommendation to execution, providing real-time visibility into system health and historical routing efficiency.

Installation

To integrate this capability into your agent environment, utilize the OpenClaw CLI package manager. Execute the following command in your terminal: clawhub install openclaw/skills/skills/adjusternwachukwu-bot/edge-router Once installed, ensure your environment variables reflect the intended base URL for the API (defaulting to https://edge-router.gpupulse.dev/api/v1). Verify the installation by querying the health endpoint via an internal tool call to ensure the agent can reach the backend router successfully.

Use Cases

• Cost-Optimized Inference: Automatically route routine LLM prompts to local hardware to keep costs at zero, only failing over to cloud GPUs when local capacity is saturated.
• Automated Training Pipelines: Offload resource-intensive model fine-tuning tasks to dedicated Vast.ai GPU nodes without manual environment provisioning.
• Quantum Research: Direct specialized scientific experiments to the Wukong 72Q backend using the designated quantum task type.
• Infrastructure Failover: Ensure high availability by using the 'auto' routing mode, which treats the local node as a primary and cloud infrastructure as a high-performance standby.

Example Prompts

1. "Route my next batch of llama3.2 inference tasks to the cheapest available backend."
2. "Execute the training job for my sentiment model; use the cloud GPU backend and report back with the status."
3. "Check the current availability of all backends and provide a summary of the latest routing statistics."

Tips & Limitations

For optimal results, ensure your local Ollama instance is configured properly before initiating tasks, as the router prioritizes local execution by default for 'inference' types. Note that while 'auto' mode provides the most cost-effective path, it introduces a slight latency overhead during the routing decision process. The quantum backend is a specialized resource; avoid using it for standard text generation tasks to prevent unnecessary costs. Always check the /health endpoint if you experience connectivity issues, and monitor the /stats endpoint periodically to audit your monthly compute spend.