Gpu

My monitoring stack used to be five containers: Prometheus, Grafana, node-exporter, cadvisor, and an uptime checker — plus a paid UptimeRobot plan for the “is the house on fire” pings. Total maintenance burden: a weekend to set up, and a Sunday every few months when something in the pipeline silently died. homelab-monitor replaces all of it with one container. Pure Python + Flask, no agents, no Prometheus, no cloud. It does the thing most dashboards skip: per-container GPU/VRAM attribution — which container is actually squatting on your VRAM — plus power-cost tracking, multi-machine monitoring over SSH, and built-in uptime checks with push alerts. ...