glintbar 1.0.0

A slim always-on-top hardware monitor that lives in the empty part of the Windows taskbar.

GlintBar

A small always-on-top hardware monitor that sits in the empty part of your Windows taskbar (or docked at the top/bottom). It shows live CPU, RAM, GPU, disk and network so you can keep half an eye on them, and it flags thermal, clock or power spikes early, before they turn into throttling or a crash like a VIDEO_TDR_FAILURE GPU hang.

What it shows

A single row of up to nine metrics, each with a value and a 60-second sparkline. Hover a metric to pop out a bigger live graph with the session min, max and average.

Tile	Source	Watch for
GPU Temp	nvidia-smi	amber >80 °C, red >87 °C
GPU Load	nvidia-smi / perf counters	sustained 100%
VRAM	nvidia-smi	red >95 %
GPU Power	nvidia-smi	sudden dropouts under load
GPU Clock	nvidia-smi	dropping to idle mid-load can mean a TDR hang
SYS Temp	ACPI thermal zone	red >97 °C
CPU Temp	LibreHardwareMonitor	real CPU package temp (needs LHM, see below)
Fan	LibreHardwareMonitor	fan RPM (needs LHM, see below)
CPU	psutil	red >95 %
RAM	psutil	red >95 %
Network	psutil	MB/s in+out
Disk	psutil	MB/s read+write

Everything comes from first-party or open-source sources and runs without admin rights. The bar only shows tiles your machine can actually provide (details below).

Supported hardware and platforms

Windows 10/11 only. It uses Win32 APIs, so there's no macOS or Linux build.

CPU, RAM, disk and network work on any Windows PC, via psutil.

GPU support is detected at launch:

GPU	Load	VRAM	Temp	Clock	Power	Source
NVIDIA	✅	✅	✅	✅	✅	nvidia-smi (driver)
AMD / Intel / other	✅	—	—	—	—	Windows GPU perf counters
No GPU / headless	—	—	—	—	—	GPU tiles hidden

AMD and Intel GPUs get load only. Temp, clock, power and VRAM need NVIDIA. The settings panel greys out anything your hardware can't provide.

Privilege levels: it works without admin, and does more with it

GlintBar itself never needs administrator rights, and it runs fine on its own. At normal privilege you get everything except real CPU temperature and fan speed: CPU, RAM, disk, network, GPU (per the table above), and the SYS Temp tile.

The SYS Temp tile reads the ACPI thermal zone through a Windows performance counter, so it needs no admin. It's a generic zone though, so treat it as a rough system-heat reading, not the exact CPU-package sensor, and some machines block it.

Real per-core CPU temperature and fan RPM need kernel-level access, which no no-admin tool can do. For those, run a helper that already does the low-level reads with admin, and GlintBar reads from it while staying at normal privilege. It picks up the extra CPU Temp and Fan tiles automatically when one is present. Two options, whichever you already use:

HWiNFO (via shared memory): in HWiNFO, open Settings and tick Shared Memory Support. That's it. GlintBar reads the shared-memory block directly (no network at all). This is easiest if you already run HWiNFO.

LibreHardwareMonitor (open source, via a local web server): in its Options menu turn on Remote Web Server (default port 8085). GlintBar reads only from http://127.0.0.1:8085 on your own machine, never the internet. Use GLINTBAR_LHM_PORT if you change the port.

Either way, start the helper before GlintBar (or restart GlintBar) and the tiles appear on their own. HWiNFO is preferred if both are running.

Requirements

Python 3.8+ and the WebView2 runtime, which already ships with Windows 11. No admin, no compiled binaries, just Python source. The only dependencies are pywebview and psutil, installed automatically.

Install and run

The simplest one-command install (isolated, adds a glintbar command):

pipx install git+https://github.com/paone9/GlintBar
glintbar

Once it is released on PyPI this becomes pip install glintbar.

From a clone, for hacking on it:

pip install -r requirements.txt
python -m glintbar

For a launch with no console window, double-click start_glintbar.vbs (or start_glintbar.cmd if Windows Script Host is blocked). To start it on login, drop a shortcut to the glintbar command (or to start_glintbar.vbs) into shell:startup (Win+R, then shell:startup).

Controls sit on the right: the gear opens settings, ● toggles CSV logging (it turns red while recording), ▤ opens the log folder, and ✕ closes.

Settings and logs live in %LOCALAPPDATA%\GlintBar (config.json and logs\), so they survive reinstalls and stay out of the install folder.

Settings

Open the gear to change:

• Metrics: pick the ones you want, with presets for Essentials, GPU focus, or All. The bar rebuilds and resizes to fit only what you keep.
• Size: Compact, Normal, or Large.
• Align: Left, Center, or Right within the taskbar gap.
• Sparklines, critical alerts, and alert sound: on or off.

Your choices are saved to config.json in %LOCALAPPDATA%\GlintBar.

Placement

Set DOCK at the top of glintbar/monitor.py:

• "taskbar" (default) floats a topmost bar over the empty part of the taskbar, between your app buttons and the system tray. It's a normal top-level window that sits above the Windows 11 taskbar's input layer, so its hovers and clicks work and it uses no extra screen space. A small background thread keeps it fitted to the gap and on top as apps open and close.
• "bottom" is a thin full-width strip just above the taskbar.
• "top" is a thin full-width strip at the top of the screen.

It lives on your primary monitor's taskbar (the one Windows treats as primary), and expects that taskbar along the bottom. On a multi-monitor setup with different scaling per monitor it stays correctly placed and sized, since the app is per-monitor-DPI aware. Only one copy runs at a time; launching it again just hands back to the running one. Letting you pick a different monitor is on the roadmap.

When a video, game, or slideshow goes fullscreen on the same screen, the bar hides itself so it's not in the way, and it comes back when you exit fullscreen.

Hover to expand

Hover a metric and a bigger popup appears just above it with a full 60-second graph, the current value, and min/max/avg. It disappears when you move off.

Critical alerts

When a metric crosses a critical threshold (GPU temp over 87 °C, or CPU, RAM or VRAM over 95%), an extra tile appears at the end of the bar. It's added rather than swapped in, so your other tiles don't move, and it stays until things settle (about two seconds to appear, with a short cooldown before it clears). There's an optional beep on a new alert so you notice without watching. Both are toggles in settings.

Away watch

If you step away and the machine keeps working hard, GlintBar figures out what was responsible and tells you when you get back. It counts you as away as soon as the screen is locked, or after a few minutes with no keyboard or mouse activity. While you're away and the CPU stays high, it records which processes are behind it. When you return it shows a short summary (how long you were away, the peak CPU and temperature, and the busiest processes) and appends a line to logs/away.csv. If nothing unusual happened, it stays quiet.

Toggle it in settings. The idle delay and CPU threshold are in config.json (away_after_min, away_cpu_pct), defaulting to 5 minutes and 25%.

CSV logging

Press ● to write one row per second to logs/glintbar_<timestamp>.csv with every metric. Open it in pandas or Excel and line it up against a workload timeline to see how throttling, thermal drift or power dropouts track with load.

import pandas as pd
df = pd.read_csv("logs/glintbar_<timestamp>.csv", parse_dates=["timestamp"])
df.set_index("timestamp")[["gpu_temp", "gpu_clock", "gpu_power"]].plot()

Notes

• It adapts to any resolution and display scaling; DPI is read at runtime.
• Light footprint: CPU, RAM, disk and network update every second, while the GPU (which means spawning nvidia-smi) is polled every two seconds and cached in between, so the monitor itself stays cheap.
• Thresholds live in the META map at the top of glintbar/ui.html, so they're easy to change.
• Ideas that aren't built yet (attention rotation, anomaly detection, toast notifications, AMD/Intel temperatures) are written up in ROADMAP.md.

Security

GlintBar is meant to be easy to review. It's plain Python with no binaries and no obfuscation, it needs no admin, and it makes no network connections at all: no telemetry, no update checks, nothing. It only reads system metrics and writes config.json and logs inside its own folder.

CI runs ruff and bandit on every push, and a CodeQL workflow runs semantic static analysis of the Python, the UI JavaScript, and the workflows themselves. GitHub Actions are pinned to commit SHAs and the two dependencies are pinned to their tested versions, so the build can't shift under you. An OpenSSF Scorecard grades the repo's security posture (badge above). See the repo's Security tab, and SECURITY.md for exactly what the app touches.

License

MIT. See LICENSE.