GPU Fleet Monitor

Monitor GPU and CPU usage across your SSH hosts directly from Raycast. See which machines are free, busy, or offline at a glance, then connect in one keystroke via your preferred terminal or editor.

Features

• Live GPU/CPU monitoring -- probes all SSH hosts in parallel and shows real-time utilization, memory usage, and top process working directories
• Quick Connect -- instantly SSH into the best available free GPU host
• Tmux session management -- list and attach to remote tmux sessions
• Add Host -- parse an SSH connection string and add it to ~/.ssh/config
• Multiple terminals -- Ghostty, iTerm, or macOS Terminal
• Multiple editors -- Cursor or VS Code (remote SSH)
• Work/Personal classification -- filter hosts by category using name patterns or SSH identity files

Setup

The extension reads hosts from your ~/.ssh/config file. No additional API keys or accounts are required.

Preferences

Preference	Description
Terminal Application	Choose between Ghostty, iTerm, or macOS Terminal
Editor Application	Choose between Cursor or VS Code
Work Host Patterns	Comma-separated glob patterns (e.g. gpu-*, ml-*) to classify hosts as work
Personal Host Patterns	Glob patterns to classify hosts as personal
Work Identity Files	SSH identity file paths (e.g. ~/.ssh/work_key). Hosts using these keys are classified as work
Personal Identity Files	Identity file paths for personal host classification
Excluded Hosts	Hostnames to hide from the fleet list (github.com is always excluded)
Default Identity File	Identity file added to new hosts created via the Add Host command
Default View	Show work, personal, or all hosts by default
SSH Timeout	Connection timeout in seconds (default: 4)
Refresh Interval	How often to re-probe hosts in seconds (default: 10)

How It Works

The extension SSHes into each host with BatchMode=yes and runs a lightweight script that queries nvidia-smi for GPU stats and /proc/stat for CPU usage. A host is considered free if GPU utilization is under 1% and GPU memory usage is under 3%.

Requirements

• macOS
• SSH access to your GPU hosts (key-based auth with BatchMode=yes)
• nvidia-smi available on remote hosts (for GPU monitoring)