A comprehensive system monitor for Linux desktops. It combines the depth of
btop and nvtop, the process management of GNOME System Monitor, and the
hardware inventory of inxi, behind one modern Qt interface.
Everything stays on your machine. No telemetry, no analytics, no update check, no network access unless you explicitly ask for a latency probe.
The dashboard. Every widget can be reordered, resized or swapped out — there are 15 to choose from.
| Page | Highlights |
|---|---|
| Dashboard | Customisable widget grid — reorder, resize, choose from 15 widgets |
| Processor | Per-core load and frequency, SMT topology, cache, governor, turbo state, RAPL power, full CPU-time breakdown, core heatmap |
| Memory | Composition bar, kernel accounting (slab, page tables, dirty, commit), swap areas, zram compression ratios, huge pages, PSI pressure stalls |
| Graphics | NVIDIA, AMD and Intel in one view — utilisation, VRAM, clocks, power, throttle reasons, per-process VRAM, OpenGL/Vulkan stack |
| Storage | Devices, partitions, filesystems, per-device IOPS, queue depth, latency, SMART health and NVMe endurance |
| Network | Per-interface throughput, Wi-Fi signal, sockets with process attribution, routing, DNS, on-demand latency probe |
| Processes | Table and tree views, inspector (threads, open files, environment), priority and affinity control, terminate/kill with confirmation |
| Services | systemd units, dependencies, boot timing breakdown, start/stop/enable via polkit |
| Sensors | Every hwmon chip — temperatures, fans, voltages, currents, power |
| Battery | Charge, health versus design capacity, wear, cycles, power flow |
| System | Distribution, kernel, firmware, Secure Boot, DIMMs, PCI/USB/audio/input devices, displays |
| Logs | Journal viewer with filtering, search, bookmarks and export |
| Diagnostics | 16 health checks, each with evidence and a recommended action |
| History | Recorded metrics, named sessions, CSV/JSON export |
| Alerts | Threshold rules with sustain windows and hysteresis |
| Settings | Theme, per-domain sampling intervals, retention, privacy, notifications |
Absent data is explained, never faked. Most monitors show 0 or a blank
when a metric is unavailable. This one returns a typed Unavailable value
carrying the reason, which the interface renders as an em dash with the
explanation in the tooltip:
Insufficient permissions: energy_uj is root-only on this kernel
You always know whether a number is zero, missing, or unmeasurable — and what you could do about it. The Diagnostics page lists every unavailable source and which are actionable.
Thresholds come from your hardware. A CPU rated to 105 °C and an NVMe controller rated to 85 °C do not share a danger point, so limits are read from the device rather than hard-coded. Alerts additionally require a condition to hold for a sustain window and use a hysteresis band when clearing, so a browser launch does not fire a "CPU critical" notification.
Monitoring is measured against itself. Every collector runs on its own cadence with its own cost instrumentation, and the Diagnostics page shows the breakdown. On the development machine the full set costs roughly 9% of one core, with the GUI thread's frame budget never exceeded.
Requires Python 3.11 or newer. Developed and tested on Python 3.14; the 3.11 floor reflects the language features used (no PEP 695 syntax) but has not been exercised in CI yet.
git clone https://github.com/Dhellss/linux-system-observatory.git
cd linux-system-observatory
python -m venv --system-site-packages .venv
.venv/bin/pip install -r requirements.txt
.venv/bin/python run.pyOr install it properly:
pip install .
observatorySee docs/INSTALL.md for distribution packages, optional tools, and the SMART permissions setup.
Every one of these is optional; the application detects what is present at start-up and explains what is missing.
| Tool | Unlocks |
|---|---|
smartmontools |
Drive health, endurance, SMART attributes |
lm_sensors |
Motherboard voltages and fan speeds |
iw |
Wi-Fi SSID, channel and link rate |
pciutils, usbutils |
Device names in the inventory |
mesa-utils, vulkan-tools |
OpenGL and Vulkan details |
polkit |
Privileged systemd unit actions |
observatory # normal launch
observatory --theme light # override the theme for one run
observatory --page CpuPage # open straight to a page
observatory --no-history # run without writing anything to disk
observatory --debug # verbose logging| Shortcut | Action |
|---|---|
Ctrl+K |
Command palette |
Ctrl+F |
Focus search |
Ctrl+1…Ctrl+9 |
Jump to page |
Ctrl+B |
Collapse sidebar |
Ctrl+P |
Pause / resume monitoring |
F5 |
Refresh current page |
Escape |
Dismiss palette, banner or search |
This is a local tool and behaves like one:
- No telemetry, analytics, crash reporting or update check.
- Nothing is written outside
$XDG_CONFIG_HOME,$XDG_DATA_HOMEand$XDG_STATE_HOME. - Process environment variables whose names suggest credentials are masked before they are ever displayed.
- The only outbound network access is the Network page's latency probe. It is disabled by default, must be enabled in Settings, targets a host you choose, and runs only when you press the button.
- The application never handles your password. Privileged systemd actions go
through
pkexec, so your desktop's own authentication dialog handles it.
app/
├── core/ Capability probing, DI container, config, units, buffers
├── models/ Immutable snapshot dataclasses + the Unavailable sentinel
├── collectors/ One per domain — pure Python, no Qt, independently testable
├── services/ Scheduler, history, alerts, diagnostics, export
├── database/ SQLite schema, connections, repositories
├── ui/
│ ├── themes/ Design tokens and generated stylesheets
│ ├── widgets/ Cards, charts, gauges, tables, navigation
│ └── pages/ One module per page
└── utils/ sysfs, PCI, hwmon and subprocess helpers
tools/ Development helpers (headless page screenshots)
tests/ 213 tests, no display server required
docs/ Architecture, development, installation
Dependencies flow one way: ui → services → collectors → models → core. The
collector layer imports no Qt at all, which is why the whole data layer can be
tested headlessly.
Read docs/ARCHITECTURE.md for the design decisions and the reasoning behind them.
.venv/bin/python -m pytest # 213 tests, no display server needed
.venv/bin/ruff check app tests # lint — clean
.venv/bin/mypy app # types — clean across 90 modules
# Render every page to a PNG for visual review, headlessly:
QT_QPA_PLATFORM=offscreen .venv/bin/python tools/screenshot_pages.py out/ darkdocs/DEVELOPMENT.md explains how to add a collector, a metric, a dashboard widget or a page. Each is a small, local change by design.
MIT. See LICENSE.