A lightweight local dashboard for monitoring a Windows AI workstation: GPU, CPU, RAM, disk, local AI services, and a long-running model-training job, all on one page.
Zero dependencies: plain Node.js (node:http) and one static HTML page.
- Live hardware stats every 2 s: GPU name, utilization, temperature, VRAM, power and fan (via
nvidia-smi), CPU %, RAM, and free disk space. - Service health for God's Eye View (port 4173), a GeoCLIP inference service (8765) and Ollama (11434), plus the Ollama models loaded and their VRAM.
- One-click restarts for GeoCLIP and Ollama (
POST /api/restart/:service, accepted only from localhost). - Warnings when the GPU passes 78 °C, VRAM passes 95%, RAM passes 90%, or a service goes down.
- Freeze/stall detective: keeps an hour of history and logs events (event-loop stalls, GPU temperature or utilization spikes, RAM pressure, GPU driver resets from the Windows event log) together with the top CPU, RAM and GPU processes at that moment.
- Training panel that reads a JSON status file from the training pipeline (progress, accuracy, median error in km).
Node.js 18+ (ES modules), vanilla HTML/CSS/JS, PowerShell, and nvidia-smi. Windows only.
git clone https://github.com/masonerlinger-lab/mission-control
cd mission-control
node server.mjs # then open http://localhost:4180To run it hidden at login, point a scheduled task at run-hidden.vbs.
Optional environment variables (see .env.example): MC_HOME (where status files live, default: your home folder) and GEV_REPO (path to the God's Eye View checkout).
| Endpoint | Description |
|---|---|
GET /api/status |
Full snapshot as JSON |
POST /api/restart/geoclip and POST /api/restart/ollama |
Restart a service |