Back to catalog
HomeLab Monitor
MITSelf-hosted homelab and AI rig dashboard in one Docker container with GPU VRAM attribution, power costs, SSH polling, and a read-only MCP server.
Open-source alternative to — compare it against:
Overview
SikamikanikoBG created HomeLab Monitor after a home cluster grew from a few Linux boxes into a multi-GPU AI rig where standard dashboards could not explain which local model held VRAM or how much electricity each inference job burned. Written in Python and shipped as a single Docker container backed by SQLite, it unifies hardware telemetry, container controls, power accounting, and a read-only Model Context Protocol server without node agents.
Why HomeLab Monitor instead of Datadog
Platforms like Datadog, New Relic, Paessler PRTG, and SolarWinds price tiers around host agents, sensor counts, or metric volume—models that punish operators running GPU workstations, NAS boxes, and Raspberry Pi nodes. Even when engineers assemble Prometheus and Grafana themselves, they still stitch together multiple exporters just to see which container or LLM process pins a graphics card. HomeLab Monitor replaces that stack with one container on port 9800 that polls Linux, Raspberry Pi, and Windows machines over SSH using Python 3 or PowerShell probes.
Its integrations target local AI and homelab operations rather than enterprise APM tracing. The hub recognizes over 30 model servers—including Ollama, vLLM, llama.cpp, and Automatic1111—attributing VRAM across multi-card rigs down to the active model and caller. A built-in read-only MCP server on port 9810 exposes 19 tools so Claude, Claude Code, or local agents can query host health, disk treemaps, and GPU load, while /metrics feeds Prometheus and alerts route to Discord, Telegram, and ntfy.sh.
What it does well
- Read-only MCP server: Serves 19 inspection tools over HTTP on port
9810, letting Claude Code or any MCP client diagnose VRAM pressure, runaway logs, or container RAM usage without write access. - Cross-vendor GPU attribution: Decodes
nvidia-smithrottle reasons, clocks, and fan speeds alongside Linuxamdgpusysfs metrics and Windows GPU counters, mapping VRAM across 30+ model servers. - Agentless SSH polling: Monitors remote Linux, Raspberry Pi, and Windows hosts over SSH, enabling remote GPU tabs, systemd status, and disk treemaps with zero remote installs.