NetSpeedTray

🎉 The 2.x era. The widget docks to the taskbar itself, so it no longer vanishes behind the Start menu or system flyouts. Everything beyond the readout lives in one calm window - the Monitor - with history charts, honest per-app and per-process detail, exportable statistics, network latency, and data-usage tracking. And since 2.1.4, the portable build updates itself - verified, swapped, and relaunched without you touching a file. Read the release notes →

Install

winget install --id erez-c137.NetSpeedTray

🧪 Trying the 2.2 beta? It’s a manual download by design - the in-app updater, WinGet and the Store never offer prereleases. Follow the pinned issue for beta builds and how to go back.

Or grab the latest Setup.exe or Portable.zip directly. Both are digitally signed - no SmartScreen warnings.

Also on the Microsoft Store - the same signed installer and the same free, open-source app, one click from the Store. It’s the only NetSpeedTray on the Store; anything with a similar name isn’t this project.

Requirements: Windows 10 or 11 (64-bit). The widget needs no admin rights. CPU/GPU temperatures on some hardware need an optional helper - see below.

Deploying to several PCs? Silent install, per-machine and per-user details, the update policy and every network connection the app makes are in DEPLOYING.md.

See it in action

The widget lives on your taskbar - at its simplest, just your up & down speeds:

  ➕ Show more on the taskbar - mini-graph, CPU/GPU/RAM, temps & power

	+ a live mini-graph
	+ CPU / GPU / RAM, temps
	everything at once

Double-click it for the Monitor - the whole story behind the numbers:

_{Overview. Live tiles + adaptive sparklines, a network hero with inline latency, a data-usage card, and the apps with the most connections - every card drills into detail.}

↓ Click to explore the rest ↓

  🌐 Network - history graph + honest per-app connections

_{Up/down history over a live per-app connection list - how many each program holds, how many are active, and the distinct hosts it reaches - with a per-NIC filter. No fake per-app "speeds": Windows can't attribute network bytes per app without a driver, so nothing is dressed up.}

  🖥️ Hardware - combined CPU/GPU graph + per-process usage

_{A combined CPU + GPU + RAM history graph over a live per-process CPU / RAM / GPU% list, plus a telemetry band of temperatures, power, and memory.}

  ⚙️ Settings - six pages, a live preview, and a Windows 11 feel

_{Appearance - fonts, colors, and arrow-style presets, with a live widget preview at the bottom that updates as you change things.}

_{General - language, update rate, startup and update checks, and configurable double-/middle-click actions. Six pages in all: General · Widget · Appearance · Hardware · Network · Advanced.}

Why NetSpeedTray?

Windows has shown network usage in Task Manager since forever - but you have to open Task Manager to see it, and it’s gone the moment you click away. That’s the gap I wanted to fill on my own machine, and I couldn’t find a widget I trusted to do it cleanly.

So I built NetSpeedTray: live up/down speeds, CPU and GPU utilization, temperatures, and power draw - all sitting in your taskbar where you can glance at them without breaking your flow. No browser tab open, no third process to babysit, no telemetry, no ads, no tracking. Built with Python and PyQt6, and signed by the SignPath Foundation so it installs cleanly.

What You Get

🌐 Network on your taskbar. Live upload and download speeds, updated every second by default. Auto-detects your primary internet connection or lets you pick specific adapters. Color-code thresholds so heavy traffic stands out at a glance. In 2.0 the readout sits inside the taskbar’s layer - it stays put through the Start menu, Quick Settings, and the tray overflow instead of disappearing.

🖥️ Hardware stats too. CPU and GPU utilization beside your network speeds, with optional temperature, power (Watts), RAM, and VRAM. Vendor-agnostic GPU support - NVIDIA, AMD, and Intel. (When do temps need a helper? →)

📊 The Monitor. One window, three tabs - Overview, Network, Hardware - with history charts, per-app connections, per-process resource use, and statistics you can export to CSV/JSON. Double-click the widget to open it.

📈 Usage & data caps. Daily / weekly / monthly totals, a settable monthly cap with a billing reset day, and opt-in 80% / 100% alerts - shown as a quiet flyout above the widget, no extra tray icon.

🩺 Network latency. An opt-in probe shows a plain-word verdict - Internet: Good / OK / Slow - with the milliseconds as quiet subtext. Defaults to your gateway (stays on your LAN).

🎨 Make it yours. Layout modes, fonts, decimals, arrow styles, color thresholds, a mini-graph overlay, light/dark auto-theming, and a live preview in Settings that shows the exact effect before you commit.

🔒 Trustworthy by design. Open source under GPLv3. Signed installer, zero ads/tracking/telemetry, logs auto-redacted of paths/IPs/MACs/hostnames/adapter names, and an in-app updater that verifies the signature before running anything. And it stays free: GPLv3 means that even if I lose interest or get hit by a bus, the code can’t be bought, closed, or paywalled out from under you - anyone can fork it.

The Monitor

Double-click the widget (or pick Monitor from the right-click menu) to open it. One window replaces the two older, separate Graph and App Activity windows. Three tabs share a single graph engine and timeline - and the Overview is intentionally chart-free, so a quick glance never loads the heavy plotting libraries.

• 🏠 Overview - the at-a-glance control center. Live tiles for network, CPU, GPU, RAM, and VRAM, each with a sparkline that adapts its scale so low-but-varying activity reads in detail. A network “hero” card shows download and upload as co-equal headline numbers with an inline latency read, a context strip across the top (session uptime, period totals, CPU + GPU power), a Data usage card, and a Most connections list (apps ranked by live connections - Windows can’t attribute bandwidth to an app without admin). Every card is clickable and drills into detail.

• 🌐 Network - history + who’s actually talking. The history graph over a per-app connection list: how many live connections each program holds, how many are active, and the distinct remote hosts it’s reaching. Honest by design - Windows can’t attribute network bytes per app without a kernel driver, so nothing is dressed up as a per-app “speed.” A per-NIC filter scopes both the graph and the totals.

• 🖥️ Hardware - the system, in depth. A combined CPU + GPU history graph (with separate-axis and one-at-a-time modes, optional smoothing, and a fixed/auto y-axis) over a live per-process CPU / RAM / GPU% list, plus a telemetry band of temperatures, power, and memory.

• 📐 Honest statistics & export. Click any card for a Statistics sheet: the real distribution (min / avg / peak, with percentiles shown only where they’re exact), peak vs off-peak, throttle and connection-drop counts. Copy or Export the window to a CSV + JSON bundle - and there’s a headless NetSpeedTray.exe --export-csv for scripted/automation use.

• ⏯️ Live / Pause & a Windows 11 feel. Freeze a chart to read a moment, then resume. Native dark title bar and rounded corners, a Fluent tab strip, remembered size/position/last-tab, a default size that fits the whole Overview without scrolling, and full keyboard navigation (Tab to the cards and tabs, Enter/Space to drill in).

Hardware Monitoring - do I need LibreHardwareMonitor?

Utilization is always free. CPU, GPU, RAM, and VRAM usage read natively through Windows performance counters and psutil - no admin, no helper, all vendors. That part just works.

Temperatures and power are the nuanced bit. NetSpeedTray tries native sources first and only falls back to LibreHardwareMonitor (LHM) when it has to. Whether you need LHM depends on your hardware:

In plain terms: on a typical NVIDIA + Intel PC you’ll see most temps and power without installing anything. If you have an AMD CPU, accurate CPU temperature and power realistically need LHM. To enable it, install LibreHardwareMonitor, run it as Administrator, and turn on Options → Remote Web Server (v0.9.5 and newer; port 8085) - NetSpeedTray detects it automatically and temps and power across all vendors light up. Leave the web server’s authentication off and let it listen on all interfaces (NetSpeedTray only ever asks 127.0.0.1). On a PC with more than one GPU, the GPU temperature and power come from the card LHM ranks first (NVIDIA, then a discrete AMD, then Intel), while GPU usage and VRAM come from Windows’ counters, which can describe a different adapter (#236).

NetSpeedTray itself never runs as admin and never ships a kernel driver. Reading CPU/GPU die temperatures requires a driver that Windows doesn’t expose to normal apps; LHM provides that driver and publishes the values, which NetSpeedTray simply reads - from LHM’s local web server on v0.9.5 and newer (127.0.0.1 only, never through a proxy), or over WMI on v0.9.4 and older. Your choice, fully optional.

Local metrics endpoint (Prometheus / Grafana)

Off by default. Turn it on in Settings → Advanced → Integrations, and NetSpeedTray serves Prometheus text format at http://127.0.0.1:9184/metrics (the port is configurable). It answers this PC only - it binds the loopback address and refuses everything else - and it never sends anything anywhere.

What’s in it: live download/upload rates (the widget’s numbers and the raw measured ones, total and per monitored adapter), the data-usage odometer for the current period and since it began, the data cap if you set one, router (and, if you enabled it, internet) latency and loss, and CPU/GPU/RAM/VRAM use for whatever you already monitor. Names follow Prometheus conventions (netspeedtray_*, bytes, seconds, _total counters).

# prometheus.yml - Prometheus running on the same PC
scrape_configs:
  - job_name: netspeedtray
    scrape_interval: 15s
    static_configs:
      - targets: ["127.0.0.1:9184"]

rate(netspeedtray_period_receive_bytes_total[5m]) * 8   # download, bits/s, from the odometer
netspeedtray_period_receive_bytes_total / netspeedtray_data_cap_bytes   # share of your cap used

Grafana reads it through that Prometheus data source. Home Assistant, or a Prometheus/Grafana on another machine, can’t reach 127.0.0.1 on this PC - by design. Run a small agent here (Prometheus in agent mode, or Grafana Alloy) that scrapes the endpoint and remote_writes to your server.

The response includes your network adapter names as labels, so anything running on this PC can read them while it’s on. Any local program, or a web page by probing, can also tell the port is open; web pages are refused before they count against the request limit, but a local program could use it up. See the Privacy Policy.

Quick Start

1. Install with the Winget command above (or the installer).
2. Right-click the widget - open Settings or the Monitor, Pause/Resume monitoring, or Exit. The top of the menu shows your data used today / this month at a glance.
3. Double-click the widget to open the Monitor (history, per-app activity, hardware, statistics).
4. Drag it to reposition - grab the widget and drag it along the taskbar to wherever you like; it remembers the spot. Turn on Free Move (No Snapping) in Settings to place it anywhere, including off the taskbar.
5. Want temperatures? Most NVIDIA/Intel systems show them out of the box. For full coverage (or any AMD CPU), install and run LibreHardwareMonitor as Administrator and turn on its Options → Remote Web Server.

FAQ

How is this different from Task Manager? Task Manager shows network speed too - but you have to open it, and it disappears the moment you click away. NetSpeedTray lives permanently on your taskbar, live every second, in the corner of your eye while you work.

Will it slow down my PC? Idle RAM is about 50 MB - that’s the number Windows Task Manager shows you (its full working set, which counts the shared Windows/Qt DLLs it maps, is ~90 MB), and CPU sits near zero between polls (~0.1%). The Monitor’s Overview tab is deliberately chart-free, so glancing at it stays light; opening the Network or Hardware charts loads the plotting library for the session, which adds roughly 50-60 MB while the Monitor is open - close the Monitor and it eases back down. The widget polls every ~1 second using the same Windows APIs Task Manager uses.

Do I need LibreHardwareMonitor? Often no - see the hardware table above. Usage stats and NVIDIA/Intel temps & power work natively; LHM (run as admin, with its Remote Web Server on) is the universal fallback and the realistic way to get AMD CPU temperature and power. The widget never asks for admin itself.

Can I set a data limit and get warned? Yes - Settings → Network → Data usage, or the tray’s Data cap…. Set a monthly cap, a billing reset day, and opt-in 80% / 100% alerts. The alert is a quiet flyout above the widget; the tray shows live progress. The cap counts what the Monitoring mode counts, on this PC only: Auto meters just the adapter your traffic is routed through (with a VPN connected that is usually the VPN’s, so its overhead and any traffic that bypasses it are left out), while Physical Adapters Only adds up every real adapter and leaves VPN, tunnel and virtual adapters out so nothing is counted twice - the closer match to your provider’s bill.

Does the latency check phone home? Only if you ask it to. The default target is your own gateway, which never leaves your LAN - a burst of four small pings every 3 seconds, about 1.3 packets a second. Pinging a public host is a separate, explicit opt-in where you choose the host. The Monitor’s Check connection button sends 20 pings to your router, and to the public host only if that opt-in is on or you tick the box that names it, just for that check. Nothing else about your network is sent anywhere - there’s no telemetry of any kind.

Why does the network name need “Location” access? The Wi-Fi band (2.4G / 5G) shows on any PC with no permission. The network name (SSID) is different: since Windows 11, Windows only hands the SSID to apps that have Location access - a Windows privacy gate, the same one your browser hits for Wi-Fi features. It is not GPS and not tracking: NetSpeedTray never uses your position, reads the name locally only to show it on the widget, and never stores or transmits it (see the Privacy Policy). Prefer not to? Leave “Network name” off and use the band - it needs nothing. The first time the name is blocked, NetSpeedTray shows a one-time explainer with a shortcut straight to the Windows Location setting.

Where does it store data and settings? %APPDATA%\NetSpeedTray\ - NetSpeedTray_Config.json for settings, speed_history.db (SQLite) for network/hardware history, and rotated logs. Everything stays local. Settings has an Export Support Bundle button that auto-redacts paths, IPs, hostnames, MACs, and adapter names for clean bug reports. The database’s schema, example queries, and how to restore a backup are all documented in DATABASE.md.

Does it work on Windows 7 or 8? Officially Windows 10 and 11 (64-bit). The taskbar APIs changed enough after Win8 that older versions aren’t tested. It might work; it’s not supported.

SmartScreen says “unknown publisher” - is that a problem? You probably grabbed an unsigned dev build. The official releases (Setup.exe / Portable.zip) are signed by the SignPath Foundation and shouldn’t trip SmartScreen.

💛 Support NetSpeedTray

NetSpeedTray is free, open source, and always will be - no ads, no telemetry, no “Pro” tier, no upsells, and no plans to ever monetize it. It’s built and maintained by one person, in spare time, because I wanted it on my own taskbar and figured others might too.

I’m not doing this for the money - but I’d be lying if I said a little support doesn’t help. If NetSpeedTray has earned a permanent spot on your taskbar, a one-time tip (or a few dollars a month) goes directly to the time it takes to fix bugs, ship features, and answer issues. When it’s one person doing all of it, even the price of a coffee genuinely makes a difference.

No pressure, ever - the app is never paywalled and never will be. If money isn’t an option, three free things help just as much:

• ⭐ Star the repo - one click, and it genuinely helps others find the project.
• 🐛 File a good bug report - use the template and attach a Support Bundle; it saves me real triage time.
• 🌍 Improve a translation - German, Spanish and Hebrew especially could use a native-speaker review. See TRANSLATORS.md.

Translators

NetSpeedTray’s UI lives in 17 languages thanks to the people below. Each translation is real time and care - if you use NetSpeedTray in your language, please give them a star and a thank-you.

See TRANSLATORS.md for how translation works and how to contribute fixes or new languages. Even one-line phrasing corrections are valuable.

# 1. Clone
git clone https://github.com/erez-c137/NetSpeedTray.git
cd NetSpeedTray

# 2. Virtual environment (PowerShell)
python -m venv .venv
.\.venv\Scripts\Activate.ps1

# 3. Dependencies
pip install -r dev-requirements.txt

# 4. Run from source
python src/monitor.py

# 5. Run the tests
pytest -v

# 6. Build
.\build\build.bat            # full signed installer + portable zip
.\build\build-exe-only.bat   # exe only (faster iteration)

Contributing

Issues and pull requests welcome. Use the bug report template for bugs and the feature request template for ideas. Even small things - a typo, a clearer error message, a translation polish - are very welcome.

Sponsors

Free Windows code signing for open source by SignPath.io, certificate by the SignPath Foundation.

Every release you download from this repo is signed end-to-end - no SmartScreen warnings, no “unknown publisher,” and a verifiable chain that the binary hasn’t been tampered with since it left CI. SignPath donates this to qualifying open-source projects, and NetSpeedTray couldn’t ship signed builds without them.

Thanks

• Translations by the community - see Translators above.
• Built on PyQt6, matplotlib, psutil, numpy, pywin32, LibreHardwareMonitor (when available), and Inno Setup.

License

NetSpeedTray is licensed under the GNU GPL v3.0.