Monitor Agent

Lightweight host metrics agent for Monitor.

Install (Linux)

curl -fsSL https://raw.githubusercontent.com/RealFascinated/Monitor-Agent/master/install.sh | sudo bash -s -- install YOUR_INGEST_TOKEN

Windows

Run this from an Administrator PowerShell to download and install the agent as a background service:

Set-ExecutionPolicy Bypass -Scope Process -Force; & ([ScriptBlock]::Create((iwr https://raw.githubusercontent.com/RealFascinated/Monitor-Agent/master/install.ps1 -UseBasicParsing).Content)) install YOUR_INGEST_TOKEN

To uninstall:

Set-ExecutionPolicy Bypass -Scope Process -Force; & ([ScriptBlock]::Create((iwr https://raw.githubusercontent.com/RealFascinated/Monitor-Agent/master/install.ps1 -UseBasicParsing).Content)) uninstall

This writes config.yml to %ProgramData%\MonitorAgent, registers a monitor-agent service via NSSM, and logs to agent.log in the same folder.

Note: The agent must run as Administrator so hardware sensors (including GPU power) can be read.

Building from source: (save install.ps1 locally or use the one-liner above with -BinaryPath)

.\scripts\build-lhm.ps1
go build -tags lhmbundle -o monitor-agent.exe ./cmd/main.go
.\install.ps1 install YOUR_INGEST_TOKEN -BinaryPath .\monitor-agent.exe

The Windows build bundles LibreHardwareMonitorLib for CPU, memory, temperatures, and GPU sensors (see licenses/LibreHardwareMonitor/NOTICE.md).

Configuration

A config file is optional. Provide settings in config.yml (see config-example.yml) or entirely via environment variables. Environment variables override values from the file.

Each YAML key maps to an environment variable: MONITOR_ + the key in uppercase (e.g. ingest_token → MONITOR_INGEST_TOKEN). New fields added to config-example.yml follow the same rule automatically.

Other runtime variables: MONITOR_HOST_ROOT (host root bind mount prefix for disk metrics in containers), MONITOR_LOG_LEVEL (debug, info, warn, error), MONITOR_PROFILE_COLLECT=1 (log per-phase collection timings at debug level).

Boolean config env vars accept true/false, 1/0, yes/no, or on/off.

If config.yml is missing and MONITOR_CONFIG_FILE is not set, the agent starts using environment variables only. If MONITOR_CONFIG_FILE points at a path, that file must exist.

Print one payload (debug)

Print one metrics payload as indented JSON to stdout (no ingest token required):

./monitor-agent print

Logs still go to stderr.

GPU metrics

Each GPU is reported in gpuMetrics with a stable deviceId (16-character hex hash of the platform identifier), plus name, vendor, and optional usage, VRAM, temperature, and power fields.

Set enable_gpu: false or MONITOR_ENABLE_GPU=false to disable.

Unraid

Install a Docker template on Unraid:

# AMD / Intel GPU, or no discrete NVIDIA GPU
wget -O /boot/config/plugins/dockerMan/templates-user/monitor-agent.xml \
  https://raw.githubusercontent.com/RealFascinated/Monitor-Agent/master/unraid/monitor-agent.xml

# NVIDIA GPU (requires the Unraid NVIDIA Driver plugin)
wget -O /boot/config/plugins/dockerMan/templates-user/monitor-agent-nvidia.xml \
  https://raw.githubusercontent.com/RealFascinated/Monitor-Agent/master/unraid/monitor-agent-nvidia.xml

Then open Docker → Add Container, choose monitor-agent or monitor-agent-nvidia, enter your Ingest Token, and apply. Defaults mount the host /proc, /sys, /dev, array root at /host, and the Docker socket for full metrics on Unraid (including /mnt/* shares and ZFS).

Docker

On Linux, container CPU and memory are collected from cgroup sysfs (/sys/fs/cgroup) when available — the same approach Netdata uses. Mount the host cgroup filesystem and optionally /var/run/docker.sock for human-readable container names. The agent falls back to docker stats only when no Docker cgroups are found.

Images are published to GitHub Container Registry on each agent/v* release tag:

ghcr.io/realfascinated/monitor-agent

The default image does not include nvidia-smi. The NVIDIA image uses an Ubuntu + CUDA base (glibc); run with gpus: all so the toolkit injects nvidia-smi and host-matched driver libraries.

Build locally:

docker build -t monitor-agent .
docker build -f Dockerfile.nvidia -t monitor-agent:nvidia .

Example docker-compose.yml (default — AMD / Intel / no discrete NVIDIA)

services:
  monitor-agent:
    image: ghcr.io/realfascinated/monitor-agent:latest
    container_name: monitor-agent
    restart: unless-stopped
    privileged: true
    pid: host
    network_mode: host
    environment:
      MONITOR_CONFIG_FILE: "-"
      MONITOR_INGEST_TOKEN: your-ingest-token
      MONITOR_API_ENDPOINT: https://monitor.fascinated.cc/api/v1/servers/ingest
      MONITOR_PUSH_SCHEDULE: "*/15 * * * * *"
      MONITOR_ENABLE_DOCKER: "true"
      MONITOR_ENABLE_GPU: "true"
      MONITOR_HOST_ROOT: /host
    volumes:
      - /:/host:ro,rslave
      - /proc:/proc:ro
      - /sys:/sys:ro
      - /dev:/dev:ro
      - /etc/os-release:/etc/os-release:ro
      - /var/run/docker.sock:/var/run/docker.sock:ro

Example docker-compose.yml (NVIDIA)

Same as above, but use the NVIDIA image and pass GPUs:

services:
  monitor-agent:
    image: ghcr.io/realfascinated/monitor-agent:nvidia
    gpus: all
    # ... same privileged, pid, network_mode, environment, and volumes as above

Alternatively, mount a config file instead of using env vars:

    environment:
      MONITOR_CONFIG_FILE: /etc/monitor-agent/config.yml
    volumes:
      - ./config.yml:/etc/monitor-agent/config.yml:ro

Releases

Tagged releases use the prefix agent/vMAJOR.MINOR.PATCH (for example agent/v2.0.1). Pushing a tag builds GitHub release binaries and publishes both Docker images (default and -nvidia) to GHCR.