Omarchy Quattro menubar widget for tracking GPU usage of a remote host via a Netdata host. Useful when running local AI on another dedicated machine. Shows GPU utilization in the menubar and a historical chart in a popup. Omarchy theme support.
- Bar pill with a status mark and the configured hostname
- Configurable LED matrix in the pill (default 5×5): each column is one refresh, newest on the right, height is GPU utilization with a green → yellow → red gradient
- Click the pill for a theme-following popup with current usage, a historical GPU utilization chart, and smaller GPU temperature, memory (used), and power draw charts locked to the same time window
- Scroll to zoom the time window; drag to pan
- Preset chips: 3D, 2D, 24H, 6H, 3H, 1H, and Live
- Colors follow the active Omarchy theme (
Color/Style) - Status mark states:
- Grey square outline when polling is stopped
- Configurable LED matrix (default 5×5) while polling: each column is one refresh, newest on the right, height is GPU utilization
- Dots are a green (bottom) → yellow → red (top) gradient using the same 30% / 70% cuts as before
- Unlit (dim grey) dots when polling but the host is unreachable or still checking
- Green and yellow stay semantic so themes that alias
greento gold still read correctly
Default host is localhost (port 19999). The default gpu is nvidia, which uses Netdata v3 context nvidia_smi.gpu_utilization. Set gpu to amd for AMD. Leave gpu blank for the same nvidia charts. Set context / tempContext / memContext / powerContext to override any of the charts.
omarchy plugin add https://github.com/BonesGit/omarchy-netdata.git --enableThe widget lands in the right section of the bar. Update with omarchy plugin update io.github.bonesgit.omarchy-netdata. Remove with omarchy plugin remove io.github.bonesgit.omarchy-netdata.
omarchy plugin update io.github.bonesgit.omarchy-netdataomarchy plugin remove io.github.bonesgit.omarchy-netdataOmarchy refuses symlinks inside a plugin folder, so if you cloned this repo, copy it in:
mkdir -p ~/.config/omarchy/plugins/io.github.bonesgit.omarchy-netdata
rsync -a --delete --exclude .git ./ ~/.config/omarchy/plugins/io.github.bonesgit.omarchy-netdata/
omarchy plugin validate ~/.config/omarchy/plugins/io.github.bonesgit.omarchy-netdata
omarchy-shell shell rescanPlugins
omarchy plugin enable io.github.bonesgit.omarchy-netdata --section right --after omarchy.trayAfter editing this repo, rsync again and restart the shell. Live bar instances do not pick up plugin-dir copies on their own; omarchy-shell shell rescanPlugins is not enough.
rsync -a --delete --exclude .git ./ ~/.config/omarchy/plugins/io.github.bonesgit.omarchy-netdata/
omarchy plugin validate ~/.config/omarchy/plugins/io.github.bonesgit.omarchy-netdata
omarchy restart shell- Netdata agent reachable from your machine (default
localhost:19999). The plugin polls the Netdata v3 HTTP API. curl— used for live and history fetches. Omarchy already ships it.omarchy-launch-browser— opens the Netdata dashboard when you click the hostname in the popup.- Default GPU:
nvidia(nvidia_smi.gpu_utilization).amdusesamdgpu.gpu_utilization. Blankgpuis the same asnvidia. Override either chart withcontext/tempContext.
The plugin only reads metrics from the host you configure. It does not install Netdata, collect GPU data itself, or modify your Netdata configuration.
Set from the bar or shell.json:
omarchy bar set io.github.bonesgit.omarchy-netdata host myhost
omarchy bar set io.github.bonesgit.omarchy-netdata refreshSeconds 5
omarchy bar set io.github.bonesgit.omarchy-netdata retryAttempts 5
omarchy bar set io.github.bonesgit.omarchy-netdata retryWindowMinutes 120
omarchy bar set io.github.bonesgit.omarchy-netdata gpu amd
omarchy bar set io.github.bonesgit.omarchy-netdata split true
omarchy bar set io.github.bonesgit.omarchy-netdata showPower false
omarchy bar set io.github.bonesgit.omarchy-netdata matrixColumns 5
omarchy bar set io.github.bonesgit.omarchy-netdata matrixRows 5
omarchy bar set io.github.bonesgit.omarchy-netdata dashboardUrl http://localhost:19999/#menu_system_submenu_gpuhost accepts hostname, hostname:port, http://hostname:port, or an IP. Port defaults to 19999.
dashboardUrl is what the popup hostname opens in the browser. Leave it empty to use http://<host>:<port>.
retryAttempts is consecutive failed live polls at the normal cadence before the wait starts doubling (default 5). The first retryAttempts failures poll every refreshSeconds (default 5s); after that the wait doubles each poll (5s, 10s, 20s, …) and settles on a 5-minute liveness probe. A successful poll resets the counter and returns to the normal cadence.
retryWindowMinutes is the maximum time to keep retrying an unreachable host before giving up (default 120 = 2 hours, range 1–1440). The window counts from the first failure of a continuous streak — any successful poll resets it — so a host that blips back online mid-retry never gets punished. Once the window expires without a success, the poller switches to "Stopped" and stays off until you flip the switch back on. Flipping the switch or changing the host starts a fresh window.
gpu is nvidia or amd (default nvidia). Blank is the same as nvidia. That pick supplies both the utilization and temperature charts. Set context or tempContext to override one or both. If you override context and leave tempContext blank, temperature is auto-picked from the override context (so an AMD context still gets the amdgpu sensors instance).
split (default off) draws one line per GPU on all four charts instead of one averaged line. The second line uses the theme's urgent color. The bar pill and hero number show the hottest GPU. Per-GPU values appear in the tooltip and the section labels. Split mode adds Netdata's group_by=instance to the query.
matrixColumns and matrixRows size the menubar LED matrix (default 5×5). Columns are time (1–24, one per refresh, scrolling left); rows are utilization (1–9). Try 3×5 or 5×4 if 5×5 feels busy.
Under utilization there are three companion sections: temperature, GPU memory, and power draw, each locked to the same time window with its own title and current value. Defaults:
| GPU | Temperature | Memory | Power |
|---|---|---|---|
nvidia |
nvidia_smi.gpu_temperature |
nvidia_smi.gpu_frame_buffer_memory_usage (used only) |
nvidia_smi.gpu_power_draw |
amd |
system.hw.sensor.temperature.input (*amdgpu*) |
amdgpu.gpu_mem_vram_usage (used only) |
system.hw.sensor.power.input (*amdgpu*) |
Memory is always requested with dimensions=used, so only the used VRAM line is fetched and drawn. showTemp, showMem, and showPower (all default on) hide a section entirely and stop polling it.
The bar stays utilization-only. A failed poll of any companion metric never stops the widget.
allowMultiple is on, so you can put another copy on the bar later and point it at a different machine. Dual-monitor copies of the same host share one poller, so online/offline and start/stop stay in sync.
- Left click the pill: open / close
- Click the hostname in the popup: open Netdata in the browser
- Right or middle click: refresh now
- Scroll on the chart: zoom time range around the cursor (up to 3 days)
- Drag the chart: pan in time. The temperature chart stays locked to the same window.
- Stop switch (top right): start/stop automatic Netdata polling. Failed polls back off (doubling the wait) and probe every 5 minutes while offline; after
retryWindowMinutesof continuous retrying (default 2h) it auto-stops until you flip the switch back on. 3D/2D/24H/6H/3H/1H/Livechips: jump the window+/-or up / down: zoom- left / right: pan
0or Live: jump back to a live 10-minute windowr: refresh- Esc: close
manifest.json plugin contract
BarWidget.qml bar pill + panel host
Panel.qml KeyboardPanel popup
Service.qml plugin singleton; one Poller per host+charts
Poller.qml Netdata v3 polling
GpuChart.qml pan/zoom canvas
CompanionChart.qml shared section for temp / memory / power
Model.js parse / window math
This follows the Omarchy marketplace bar-widget + nested panel pattern.
More netdata metrics and visualizations coming soon.