runpod

What This Skill Does

The RunPod skill provides a comprehensive interface for managing cloud-based GPU infrastructure directly from your OpenClaw agent. It bridges the gap between local command-line operations and remote cloud resource management, allowing the AI to provision, monitor, and interact with high-performance NVIDIA GPU instances. The skill leverages the official runpodctl utility to handle the full lifecycle of pods, including creation with specific GPU models, environment image selection, and persistent volume management. Beyond simple provisioning, it facilitates complex remote development workflows by handling SSH key management, direct terminal access, and filesystem mounting via SSHFS. This enables users to treat remote GPU clouds as if they were local directories, streamlining tasks like model training, large-scale data processing, and remote development environments.

Installation

To use this skill, ensure you have the runpodctl utility installed on your system. Run brew install runpod/runpodctl/runpodctl. Once installed, configure your authentication by running runpodctl config --apiKey "your-api-key". Additionally, generate and register your SSH keys using runpodctl ssh add-key to ensure secure connectivity to your instances. If you are using custom directory paths for your SSH identity, set the RUNPOD_SSH_KEY environment variable to point to your private key file.

Use Cases

This skill is designed for AI engineers and developers who require scalable compute power. Common use cases include:

1. Spinning up ephemeral high-VRAM pods for one-off deep learning training tasks.
2. Managing persistent development environments with mounted volumes for code editing across sessions.
3. Prototyping and testing models that require specific CUDA versions via custom Docker images.
4. Providing remote debugging capabilities to the AI agent by allowing it to inspect logs and system status inside the pod via terminal.
5. Interacting with services like ComfyUI, Jupyter, or Gradio hosted on RunPod via the built-in proxy URL system.

Example Prompts

1. "Create a new RunPod instance named 'training-node' using an RTX 4090 and the latest PyTorch image with 100GB of persistent storage at /workspace."
2. "List all my currently running pods and tell me the status of the instance with ID 'rpod-12345'."
3. "Mount the filesystem for pod 'rpod-12345' so I can edit the training scripts directly from my local machine."

Tips & Limitations

Always remember to specify the --volumePath if you provide a --volumeSize during creation; the infrastructure requires this explicitly to prevent mount configuration errors. Use the provided mount_pod.sh script for seamless filesystem integration, but ensure you unmount using fusermount -u to prevent local hang-ups. Be aware that the skill handles host keys in a local isolated directory (~/.runpod/ssh/known_hosts) to keep your system SSH config clean. Note that web services are accessible via the standard RunPod proxy format, so you can often ask the agent to help you generate the correct URL to access your remote Jupyter or Gradio dashboard.