Prometheus & Grafana¶

HomeLab Monitor exposes a standard Prometheus scrape endpoint at /metrics (port 9800 by default). It reads exclusively from the in-memory snapshot that the background collector keeps fresh — no extra polling, no double-sampling.

Metrics exposed¶

Metric	Labels	Description
`homelab_gpu_vram_used_mb`	`gpu`	GPU VRAM currently used (MB)
`homelab_gpu_vram_total_mb`	`gpu`	GPU VRAM total capacity (MB)
`homelab_gpu_util_pct`	`gpu`	GPU utilisation (%)
`homelab_gpu_temp_c`	`gpu`	GPU temperature (°C)
`homelab_gpu_power_w`	`gpu`	GPU power draw (W)
`homelab_host_cpu_pct`	—	Host CPU usage (%)
`homelab_host_mem_used_pct`	—	Host memory used (%)
`homelab_host_disk_used_pct`	`mountpoint`	Disk used per mount (%)
`homelab_container_state`	`name`, `state`	1 = container is in this state
`homelab_systemd_unit_state`	`unit`, `state`	1 = unit is active, 0 = otherwise
`homelab_model_loaded_vram_mb`	`server`, `model`	VRAM used by a loaded model (MB)

Quick verification¶

curl http://<your-host-ip>:9800/metrics

Sample scrape config¶

# prometheus.yml
scrape_configs:
  - job_name: 'homelab_monitor'
    scrape_interval: 15s
    static_configs:
      - targets: ['<your-host-ip>:9800']

Grafana dashboard¶

A ready-to-import dashboard is at docs/grafana/homelab_prometheus_dashboard.json. In Grafana: Dashboards → Import → Upload JSON file, then select your Prometheus datasource. The dashboard covers GPU VRAM, utilisation, temperature, host CPU/RAM, disk usage, and model VRAM in a single view.

Note

Prometheus is entirely optional — the dashboard is fully self-contained without it. The endpoint is there for folks who already run a Prometheus/Grafana stack.