sysresource-monitor

CPU, RAM and per-drive disk usage on a small always-on-top toast that parks itself just above the system clock - and slides out of the way the moment your cursor comes near it.

🖥  System Resources                       61%
    CPU   6.2 of 8 cores idle   25°C       23%
          ▁▃▁▂█▁▂▁                             <- one bar per logical core
    RAM   6.7 GB free of 15.9 GB           58%
    CPU↑  python.exe                        7%
    RAM↑  studio64.exe   1.7 GB            11%
    C:    161.0 GB free of 350.9 GB        54%
    D:    62.2 GB free of 125.2 GB         50%
    G:    152.9 GB free of 350.9 GB        56%
    updated 23:23:24   every 5s   3 drives   % = in use

No config file, no HTTP server, no background service - a tray icon and a toast, plus a CLI for when you just want the numbers.

Install

pip install sysresource-monitor

From a checkout:

pip install -e .

Requires Python 3.11+, PySide6 for the toast and psutil for the numbers.

Use

sysresource-monitor                    # launch the tray toast
sysresource-monitor --interval 2       # ... sampling every 2 seconds
sysresource-monitor --all-mounts       # ... including mounts psutil hides
sysresource-monitor status             # print one sample as a table
sysresource-monitor status --json      # print one sample as JSON
sysresource-monitor status --swap      # include swap usage
sysresource-monitor status --watch     # redraw the table every 5 seconds
sysresource-monitor status --watch 2   # ... every 2 seconds instead
sysresource-monitor status --no-procs  # skip the top-consumer scan
sysresource-monitor status --no-temp   # skip the temperature query

Also runs as python -m sysresource_monitor [...], or on Windows via run_monitor.bat, which pins the py -3.13 launcher so a stray python on PATH can't pick an interpreter without PySide6 installed.

What it shows

Each row pairs what is spare (in words, on the left) with what is in use (the percentage, on the right). Those are opposite quantities, so the left-hand text always names itself - idle, free - and the percentage column is labelled once, in the footer, as % = in use.

The core strip earns its place on a busy machine: one pegged core reads as 12% overall on a 8-thread box, which looks calm on the CPU row and obvious on the strip.

Per-process CPU is normalised by core count, so CPU↑ 25% means a quarter of the whole machine, the way Task Manager counts it - not a quarter of one core. The terminal view prints both figures.

About the temperature

There is no portable way to read a CPU temperature, so this is best-effort and the row simply omits it when nothing will answer:

• Linux/FreeBSD - psutil.sensors_temperatures(), preferring a real package sensor (coretemp, k10temp, zenpower, cpu_thermal).
• Windows - psutil doesn't implement that call at all, and the usual substitute, MSAcpi_ThermalZoneTemperature, needs an elevated process. The fallback is the ACPI thermal-zone performance counter, which any user can read. That is a chassis/zone reading rather than the CPU die, so treat it as indicative. If you want a true die temperature, run something like LibreHardwareMonitor alongside.
• macOS - nothing without a signed helper, so the row stays hidden.

The badge in the header is the highest pressure of all of them, and it colours the whole card: green under 70%, amber from 70%, red from 90%. Each row's own percentage is coloured on the same scale, so a drive filling up stands out without you reading the numbers.

Where it sits, and how it moves

The toast anchors to the bottom-right of the work area - the desktop minus the taskbar - so on a default Windows setup it lands directly above the clock. Move the taskbar or change its size and the toast follows on the next sample.

The touch-and-go behaviour is the same as cursorhub's: bring the mouse into the toast's corner and it slides to the opposite side of the screen in a 250 ms ease-out, then slides home once the cursor leaves. It never takes focus and never eats a click - it just gets out of the way.

Refresh cadence

Three cadences, because the numbers cost wildly different amounts to read:

Sampling happens on a worker thread, so a slow mount never freezes the toast, and a tick is skipped rather than queued if the previous one is still running. Values from the slower passes stay on screen between them, so no row ever blanks.

Per-process CPU counters are primed once at the first scan, because cpu_percent() has nothing to diff against on a first call - without that, the first reading is 0.0 for every process and the "top consumer" is just whichever process happened to be enumerated first.

Unreadable mounts - empty optical drives, offline shares, mounts you can't stat

• are skipped rather than raised. A monitor that dies because a share went offline is worse than one missing a row.

Tray menu

• Refresh now - resample immediately, disks included
• Show toast / Hide toast
• Quit

Single-click or double-click the tray icon to bring the toast back.

Layout

sysresource_monitor/
    stats.py       psutil sampling + byte/percent formatting (Qt-free)
    monitor.py     the toast, the tray icon, the refresh timer
    cli.py         `status` table, --json, --watch
    __main__.py    argument dispatch

stats.py imports nothing from Qt, so sysresource-monitor status works over SSH or in a container where there's no display to put a toast on.

Licence

MIT - see LICENSE.