System Monitor for Omarchy

A compact bar widget that opens into a live system dashboard. See what is busy, what changed, and which processes are responsible — without leaving your desktop.

Resources	Activity	Processes	Settings

The dashboard

• Overview: current CPU and memory, GPU, network, storage, uptime and load averages. Switch the detailed chart between CPU, memory and GPU, with 1, 5 or 15 minutes of history and hover readouts.
• Resources: every logical CPU, package temperature, available memory, reclaimable cache, swap, GPU/VRAM, and CPU/memory/I/O pressure.
• Activity: download/upload rates, auto-scaled traffic history (hover for both rates at a moment), interface lifetime totals, and storage usage/free space for the selected mountpoint.
• Processes: live CPU and resident memory, PID, user and state. Search by name/user/PID, sort by CPU/memory/name, or pause the list to inspect it. The list keeps your scroll position as it refreshes.
• Settings: choose bar chips, bar style, units, refresh interval, interface and mountpoint. Changes use the same settings as Omarchy's plugin settings screen.

The design takes cues from btop, htop, Glances, and the Trading212 plugin. See design and measurement notes for the reasoning.

Install

omarchy plugin add https://github.com/simasrazinskas/omarchy-sysmon-plugin.git --enable

Place it with omarchy bar move io.github.simasrazinskas.sysmon, or drag it along the bar. Requires Omarchy's Quickshell-based shell. Python 3 is used only for the Processes view; other readings work without it. No pip packages, background daemon, elevated privileges or remote service are needed.

Controls

Click the bar widget to open the dashboard. The gear opens settings.

Key	Action
1–4	Overview / Resources / Activity / Processes
← / → or h / l	Previous / next view
↑ / ↓ or k / j	Scroll the view
/	Open Processes and focus search
Space / Enter	Pause/resume the process list
,	Toggle settings
Esc	Leave a text field; otherwise close the panel
Tab / Shift+Tab	Switch to the next/previous Omarchy panel

Bar and settings

Bar readings keep a fixed width, so changing numbers do not move neighboring widgets. Vertical bars stack values. Switching off every reading leaves a clickable icon, and unavailable hardware never produces a misleading zero.

Choose a bar style under Settings → Bar style:

Style	Shows
Icons + values	The default: an icon next to each reading
Values only	Readings without icons
Minimal	A single icon; the readings stay in the tooltip
Alert colours	Theme colour until a reading is high, then amber (warning) or your theme's urgent colour (critical)
Tinted	A slight, distinct hue per metric
Grouped tint	A hue family per device: CPU usage azure / CPU temp ocean blue, GPU violet / plum / orchid, network teal / seafoam, memory green, disk amber
Grouped tint, values only	Grouped tint without icons — colour already says which reading is which

Tints are mixed into your theme's bar colour rather than replacing it, so they stay subtle on light and dark themes.

Alert thresholds (warning / critical) are shared by the bar's alert style and the dashboard's meters and charts, so they never disagree:

Reading	Warning	Critical
CPU, GPU, per-core	≥75%	≥90%
Memory, VRAM	≥80%	≥92%
CPU/GPU temperature	≥75°C	≥90°C
Disk	≥85%	≥95%

Setting	Default
CPU, memory, CPU temperature, GPU, GPU temperature, network, disk chips	On
VRAM chip	Off
Bar style	Icons + values
Memory shown as	Used bytes (or percent)
Temperature	Celsius (or Fahrenheit)
Refresh interval	2 seconds; configurable from 1–30
History window / chart metric	1 minute / CPU
Network interface	Automatic: follows the default route
Disk mountpoint	/

Network auto-detection runs every 30 seconds. Switching interfaces resets rate baselines and clears old traffic from history. A named interface that doesn't exist shows "not found" rather than zero traffic. Storage always follows the configured path, not whichever drive happens to be largest. Text settings save on Enter or when you leave the field, only if the value changed; Esc restores the saved value. Sensor discovery uses hwmon names and prefers package-wide CPU readings.

Reading the numbers

• Memory: used = total − MemAvailable. Reclaimable cache is not all treated as used memory. The cache row includes Cached + SReclaimable.
• Process CPU: 100% means one fully occupied logical core. A multithreaded process can exceed 100%. CPU needs two observations; the first shows —.
• Process memory: RSS, including shared pages. Adding rows does not equal total used memory. Process names come from the kernel's comm field and can be truncated by the kernel; command arguments are not read.
• Pressure: the percentage of time with tasks waiting for CPU, memory or I/O over the kernel's recent ten-second window. Unsupported kernels show “Not supported.” See Linux PSI documentation.
• History: up to 15 minutes in memory, collected while the shell runs. It starts empty after reload. CPU/memory/GPU charts use a fixed 0–100% scale; network auto-scales. Missing readings and long sampling gaps break the line.
• Traffic totals: cumulative counters for the selected interface since it was created/reset, not monthly usage or totals since opening this panel.
• Storage: df -Pk; free space is space available to applications, which can exclude reserved blocks. Capacity updates once a minute.

Sampling and overhead

Data	Source	Frequency
CPU/per-core, memory, temperature, network, pressure, load, uptime	In-process /proc and /sys file reads	Refresh interval
GPU	nvidia-smi or sysfs backend	At least 5 seconds
Storage capacity	df -Pk	60 seconds
Processes	Small standard-library Python procfs reader	At least 2 seconds, only while Processes is open
Route / unavailable hardware retry	ip route, sensor/GPU probes	30 seconds

The history buffer is capped at 901 samples. Closing the panel unloads its views and charts. Process snapshots have a five-second timeout and never scan application files or request privileged access. Pausing freezes the displayed list while sampling continues. On multi-monitor setups, Omarchy instantiates one service per bar surface.

Hardware support

CPU, memory, network and storage work on Linux. GPU backends are selected in NVIDIA → AMD → Intel order; unavailable backends are retried so driver resets or reconnects can recover. Only the first readable GPU is displayed.

Backend	Support
NVIDIA	nvidia-smi; tested on RTX 5070 Ti
AMD	amdgpu sysfs; not verified on physical AMD hardware
Intel	gpu_busy_percent where exposed; not verified on physical Intel hardware. Most i915 systems lack this reading and show unavailable

With no readable GPU, the dashboard says so instead of drawing an empty chart. Missing sensors remain unavailable. Battery, fan speeds, per-disk I/O, process termination, process trees and persistent history are not included.

Development

./tests/run         # JS/Python regression tests + manifest validation
./tests/qml-smoke   # Native live-data checks; requires installed Omarchy/Quickshell
./tests/screenshots # Regenerate docs/*.png from live readings (~90 s)

Keep the smoke test's rendered frames with SYSMON_CAPTURE_DIR=/tmp/sysmon-preview ./tests/qml-smoke. For a quick screenshot preview, use SYSMON_WARMUP=5 ./tests/screenshots /tmp/shots.

File	Role
Panel.qml	Entry point: bar button, popup, keyboard routing, saving settings
Service.qml	Samples the machine; GPU backends live in *Backend.qml
Dashboard.qml	Popup frame: heading, tabs, active view, footer
views/	One file per tab, plus settings
components/	Shared UI: BarReadings (the bar itself), charts, cards, theme
Model.js	Bar logic: chips, styles, thresholds, compact formatting, tooltip
Metrics.js	Kernel parsing and dashboard maths
scripts/processes.py	One read-only procfs snapshot for the Processes view

Model.js and Metrics.js are pure and Node-tested.

License

MIT