runpodctl

What This Skill Does

The runpodctl skill provides a robust interface for the RunPod command-line utility, enabling seamless management of cloud-based GPU infrastructure directly through the OpenClaw agent. This skill empowers users to provision, monitor, and decommission GPU pods, serverless endpoints, network volumes, and machine learning models without ever leaving the terminal environment. By wrapping the runpodctl binary, the agent handles the heavy lifting of API communication, resource allocation, and state management, making it an essential tool for AI researchers, machine learning engineers, and developers who require scalable compute power on demand.

Installation

To install this skill, run the following command in your terminal: clawhub install openclaw/skills/skills/itamarcoh3n/runpodctl

Ensure you have your RunPod API key ready. Upon the first use, you must configure the environment by executing the configuration command: ~/.local/bin/runpodctl config set --apiKey YOUR_API_KEY You can verify your connection and check your account balance by running ~/.local/bin/runpodctl user.

Use Cases

• Scalable GPU Provisioning: Launch specific GPU instances (like the NVIDIA RTX 4090 or A100) instantly for training models or running inference.
• Serverless API Management: Deploy and update serverless inference endpoints, allowing you to scale your models based on traffic automatically.
• Infrastructure Orchestration: Create and attach persistent network volumes to your pods, ensuring your data remains available across pod restarts.
• Resource Monitoring: Regularly audit active pods, check current costs, and inspect GPU availability to optimize cloud spending.

Example Prompts

1. "List all my currently running GPU pods and tell me how much credit I have left in my account."
2. "Search for an official PyTorch template and create a new pod using an NVIDIA A100 80GB GPU."
3. "Check the current status of my serverless endpoint 'my-model-api' and increase the maximum number of workers to 10."

Tips & Limitations

• Always use the absolute path ~/.local/bin/runpodctl to ensure the agent correctly resolves the binary location.
• GPU IDs must match the string requirements exactly; use runpodctl gpu list to fetch the correct naming conventions.
• If a pod creation fails, double-check your template ID; you can use runpodctl template search to verify availability.
• Note that pod endpoints are accessible via the proxy.runpod.net domain; ensure your security group settings allow the necessary ports.