English · 简体中文
A real-time dashboard for the USB-C dock or hub plugged into your computer, served on localhost. It shows what is behind the dock, at what link speed, what just got plugged in — and where the power is going.
No dependencies. Node 16+. Reads the operating system's own device tree; it does not install a driver, a kernel extension or a background service.
npx dock-consoleOr without a dock attached — synthetic data, same interface:
npx dock-console --demo- Link strip — host → uplink → dock → port chips, so you can see what hangs off which port.
- Power chart — total draw with the CPU and GPU contributions, 1 Hz, hover for exact values.
- Load — CPU, GPU, memory, disk and network, plus battery voltage and health.
- Device table — every USB node, split into behind the dock and directly on the host, with port number, negotiated speed, VID:PID and live disk throughput.
- Event stream — hot-plug, dock connect/disconnect, ethernet link changes, AC/battery transitions.
| Windows | macOS | Linux | |
|---|---|---|---|
| Dock topology, ports, VID:PID | pnputil |
system_profiler |
sysfs |
| Negotiated link speed | inferred from the hub model | system_profiler |
…/speed |
| CPU / memory | PDH via typeperf |
top |
/proc |
| Disk / network throughput | PDH | iostat + netstat |
/proc |
| GPU utilisation and watts | nvidia-smi |
nvidia-smi |
nvidia-smi |
| CPU package watts | — | — (needs root) | RAPL |
| Whole-machine watts | modelled | measured on battery | measured on battery |
Windows does not expose a per-device or per-rail power figure to user space, so there the total is a model. Everywhere else it can be a measurement. The UI labels which one you are looking at, and never presents an estimate as a reading.
This is the part that matters in a tool like this, so it is spelled out rather than buried:
| Reading | Source | Real? |
|---|---|---|
| Device tree, port numbers, VID:PID, dock detection | OS device enumeration | measured |
| Hot-plug events | diff of consecutive enumerations | measured |
| Disk / network throughput, CPU, memory | OS performance counters | measured |
| GPU utilisation, temperature, watts | nvidia-smi |
measured |
| CPU package watts (Linux) | RAPL energy counter | measured |
| System watts on battery | battery voltage × amperage | measured |
| System watts on AC, and everywhere on Windows | base + CPU% × TDP + GPU + disk |
estimated |
The power badge on the KPI card reads either measured or estimated, and the footer says which formula is in play. A GPU line is hidden entirely when nvidia-smi is absent, rather than drawn at zero.
What it cannot tell you: which physical port carries your display signal. No mainstream OS exposes that mapping, so the channel row shows the display and says nothing about the path.
--port <n> HTTP port (default 8787)
--host <addr> bind address (default 127.0.0.1 — localhost only)
--demo synthetic dock, no hardware access
--no-open don't launch a browser
--no-power hide the power panel
--cpu-watts <n> peak CPU package power used by the estimate (default 45)
--config <file> path to a JSON config file
-h, --help
-v, --version
Copy config.example.json to config.json. Everything is optional:
{
"port": 8787,
"dock": {
"name": "My dock",
"vendorId": "2109"
},
"power": {
"baseW": 8,
"cpuIdleW": 4,
"cpuMaxW": 45
}
}power.cpuMaxW is the only value you are likely to want to change — it should be your CPU's sustained package power (PL1), not its turbo figure. dock.vendorId pins detection to one hub vendor when the automatic choice picks the wrong tree.
The UI follows your browser language (English, 简体中文) and can be switched at runtime with the button in the top right; the choice is remembered. ?lang=zh / ?lang=en force one.
src/
index.js orchestration, 1 Hz frame assembly, SSE broadcast
platform/
win32.js PowerShell collector + typeperf
darwin.js system_profiler + top + iostat + netstat + ioreg
linux.js sysfs + procfs, no child processes
model.js platform-neutral device model, dock detection, diffing
power.js source priority: battery > RAPL > model
server.js static files + SSE
public/ the dashboard (vanilla JS, canvas, no framework)
test/ parser fixtures + end-to-end smoke test
The platform layer is the only place that knows what OS it is on. Everything above it works on a normalised device shape:
{ id, parent, desc, busname, vid, pid, speed, isHub, root, physical, kinds }Linux and macOS adapters fill speed with the device's negotiated rate, which is why they can show "USB 3.2 Gen 1 · 5 Gbps" per device instead of guessing. Windows gets — there and falls back to inferring the lane from hub descriptions.
Events are emitted as a translation key plus parameters, never as pre-rendered prose, so switching language re-renders the history too.
npm test # 37 unit tests — parsers, device model, power priority
npm run smoke # boots the real server in demo mode and walks the HTTP surfaceThe Linux and macOS adapters cannot be executed on a Windows development machine, so their parsers are pure functions over file contents and are covered with fixtures. The smoke test runs on all three platforms in CI (.github/workflows/ci.yml).
- Windows has no per-device wattage. USB port power is not exposed; the total is a model and is labelled as one.
- macOS disk throughput is read+write combined.
iostatdoes not split it; the UI says so instead of inventing a direction. - macOS needs Accessibility permission for nothing, but
system_profilertakes ~1–2 s, so the device tree refreshes every 4 s there. - Port numbers are OS enumeration addresses (
Port_#0004.Hub_#0006), not the numbers silkscreened on the case. They rarely match. - The dock is auto-detected as the hub subtree with the most interesting topology (HS+SS pair, nested hubs). A dock without an internal hub cannot be distinguished from a plain device.
- Per-port current draw on Linux (
…/power_supplywhere the hub exposes it) - CSV export of the power history
- Notifications on dock disconnect / over-current
MIT. See LICENSE.
