A CLI-driven skill for deploying and managing high-performance vLLM-based Large Language Model services across multiple GPU servers.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install gpu-deploy
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install gpu-deploy using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is GPU Deploy?
GPU Deploy is a specialized tool designed to simplify the lifecycle of hosting Large Language Models using the vLLM framework. As a key part of the Openclaw Skills library, it provides developers with a structured way to manage remote GPU resources, check for hardware availability, and launch high-performance inference servers with minimal manual intervention.
By abstracting complex SSH commands and environment setups, this skill enables rapid deployment of industry-leading models like DeepSeek, Llama, and Qwen. It ensures that your AI infrastructure is always ready for production or testing environments, providing a professional-grade bridge between raw hardware and accessible AI APIs through the Openclaw Skills framework.
GPU Deploy Use Cases
- Automating the deployment of vLLM inference servers on remote hardware.
- Monitoring GPU memory and port availability before launching new LLM instances within Openclaw Skills workflows.
- Managing a fleet of GPU servers through a centralized configuration file.
- Quickly switching between different open-source models like DeepSeek-R1 and Llama 3 for testing and development.
How GPU Deploy Works
- The user defines server connection details and resource limits in a local JSON configuration file.
- The skill executes remote diagnostic checks via SSH to verify GPU health, VRAM usage, and port availability.
- A specific model is selected from a pre-configured library or a custom user-defined model entry.
- The skill initiates a remote session (leveraging tmux and conda) to serve the model using optimized vLLM parameters such as tensor parallelism.
- Active services are monitored and managed through CLI commands to ensure continuous operation and easy resource cleanup.
GPU Deploy Setup
First, ensure you have SSH access to your target servers. Create your server configuration directory and file:
mkdir -p ~/.config/gpu-deploy
touch ~/.config/gpu-deploy/servers.json
Populate servers.json with your server metadata:
{
"servers": {
"my-gpu-node": {
"host": "192.168.1.100",
"user": "ubuntu",
"gpu_count": 4,
"model_path": "/data/models"
}
},
"default_server": "my-gpu-node"
}
Verify the connection using the check command:
gpu-deploy check
GPU Deploy Data Schema & Taxonomy
The skill utilizes a configuration-first approach to manage infrastructure metadata and model parameters within the Openclaw Skills environment.
| File | Purpose | Key Metadata Fields |
|---|---|---|
servers.json |
Infrastructure Map | host, user, gpu_count, model_path |
models.json |
Model Definitions | tensor_parallel_size, max_model_len, dtype, port |
SKILL.md |
Skill definition | requires: bins (ssh), emoji: 🚀 |
GPU Deploy Advanced Features
- Multi-GPU orchestration using tensor parallelism to handle large parameter models effectively.
- Custom model definition support via models.json for hosting private or specialized weights.
- Integrated process management (ps command) to track running vLLM instances across different ports and servers.
- Automated environment activation and optimized path resolution for remote nodes using Conda and Tmux integrations.
SKILL.md
Loading
Related Openclaw Skills
Reddit VOC Lobster Pro
An advanced AI-powered Reddit Voice of Customer research engine that transforms community discussions into professional business reports.
Amazon to Shopify Sync Skill
A specialized tool for extracting Amazon product data and automatically listing it on Shopify stores with multilingual translation.
Astra Docker Workspace
Execute commands and manage files within a persistent Astra Docker container workspace.
YouTube Lecture Transcript Analyzer
A professional tool for extracting core structural insights, key arguments, and actionable steps from YouTube lecture transcripts.
New API Usage Tracker
A specialized utility for querying and monitoring usage statistics, token consumption, and remaining balances from any new-api deployment.
Xiaoai Voice Command Integration
Triggers local bash scripts when user queries begin with the Xiaoai assistant wake word.
Featured*
