English | 简体中文
Continuously track your Remote-SSH host's CPU / memory / disk / network / GPU / Docker status — as quiet yet always-at-a-glance as VS Code's native latency indicator — without breaking your coding flow.
Most similar extensions just move a full dashboard into the status bar: dense, and permanently on display. Remote Pulse takes a different approach:
| Aspect | Common approach elsewhere | Remote Pulse |
|---|---|---|
| Always-on footprint | CPU|MEM|DISK all shown flat | Just 1 icon + 1 core number by default; everything else lives in the trend panel |
| Visual tone | Value-driven coloring at all times | Each metric colored on a green/yellow/red scale independently of the others — the same visual language as VS Code's own remote-connection indicator |
| Interaction | Some require opening a sidebar | Click to pop a lightweight Webview — no persistent space used, no trace left after closing |
| Resource cost | Some poll via spawned subprocesses | Reads /proc directly; zero steady-state subprocess overhead for core metrics |
| Context awareness | Polls at the same rate whether focused or not | Automatically throttles when the window loses focus |
$(pulse) CPU 23% MEM 61% ← all normal
$(pulse) CPU 85% MEM 40% ← CPU alone crosses warning, turns yellow — memory stays default
$(warning) CPU 28% MEM 97% DISK 88% GPU 12% $(arrow-down) 240.0 KB/s $(arrow-up) 30.0 KB/s ← memory alone goes critical, turns red — the icon follows the worst of what's shown
$(debug-pause) ← paused via remotePulse.enabled; click to resume
CPU, memory, disk (fullest mount point), GPU (primary GPU only) and network are up to five independently colored status bar items — pick which ones appear via remotePulse.statusBarMetrics (CPU/memory are on by default). Network shows download and upload separately with $(arrow-down)/$(arrow-up) icons rather than one combined number, since a single figure can't tell you which direction is actually busy; each rate is padded to a fixed width so the item doesn't jitter on every refresh. A shared alert icon reflects the worst level among the ones you've enabled — network is display-only and never colors it, since throughput has no natural 0-100% scale — using VS Code's own statusBarItem.warning*/error* theme colors, so it stays legible no matter what the status bar's actual background happens to be (e.g. Remote-SSH recoloring the whole bar).
Hover any metric item for a tooltip carrying its last-2-minutes sparkline and the host it belongs to; the alert icon's tooltip explains that clicking it opens the metric picker.
Click the CPU/memory/disk/GPU/network items — or run Remote Pulse: Show Trend Chart — to open a line chart of the last 30 minutes of history, plus disk/network/GPU/Docker detail (a Webview that's destroyed on close — nothing stays resident in memory). Click a legend entry to hide that line temporarily (the axis scales stay put, so hiding a line doesn't rescale the others). The alert icon itself instead jumps straight to a multi-select picker for statusBarMetrics — VS Code's Settings UI can only render array settings as a list editor, not real checkboxes, so this command (and its trendPanelSections/trendChartMetrics counterparts, all reachable from the panel's gear icon) is the actual "check all that apply" experience.
Three settings overlap in what they can show but are independent on purpose: statusBarMetrics (cpu/memory/disk/gpu/network) picks the status bar summary, trendPanelSections (network/gpu/docker) picks which detail sections/rows show in the panel body (GPU cards, Docker table, the network row in "System" — System and Storage are always shown), and trendChartMetrics (cpu/memory/gpu/network) picks which lines get plotted in the 30-minute chart specifically — so you can, say, keep the GPU detail card visible in the panel without cluttering the chart with a GPU line, or vice versa.
- CPU: overall usage and core count (delta-based
/proc/statcalculation, not loadavg) - Memory: usage percentage and used/total (uses
MemAvailablerather thanMemFree, which better reflects what's actually available) - Disk: per-mount-point usage (virtual filesystems are filtered out automatically, and mount points that report byte-identical total/used space — the same underlying disk bind-mounted at more than one path, e.g. WSL2's
/mnt/wslg/distromirroring/— are collapsed to whichever path is shallower; shows every remaining real mount point by default, sorted by usage so the fullest one leads, or specify mount points manually to show only those). As a status bar item it reports the fullest mount point, since that's the one that will actually run out - Network: download/upload rate, always shown as two separate numbers (status bar icons, chart lines, or arrows) rather than a combined figure — merging them hides which direction is actually busy; loopback and virtual interfaces (
lo,docker0,veth*,br-*,virbr*,tap*,dummy*) are excluded, since container-to-container traffic never leaves the machine and counting it would inflate "network speed" well above what the NIC actually carries. Optionally plotted as two lines in the 30-minute chart sharing their own right-hand axis, scaled to the window's own peak since throughput has no natural 0-100% scale like CPU/memory (off by default viatrendChartMetrics, independent of whether the rate is shown as a detail row) - GPU: VRAM usage, utilization, temperature per card (requires
nvidia-smi; if it's missing the section shows why instead of silently vanishing, and the probe is retried on an interval so a WSL GPU attached after startup still shows up) - Docker: running container count plus per-container CPU/memory usage. The socket is resolved from
DOCKER_HOST(when it's aunix://URL, matching the docker CLI), then/var/run/docker.sock, then the rootless-docker/podman user sockets under$XDG_RUNTIME_DIR— so a rootless setup isn't silently excluded. Missing socket and missing permission get different hints in the panel - Threshold alerts: CPU, memory, disk and GPU each turn green/yellow/red independently as they cross the warning/critical thresholds, with an optional system notification. It fires once per crossing into the critical state (so it won't spam you) and carries two actions: jump to the trend panel, or mute alerts for 10 minutes
- History trend: a Webview line chart of the last 30 minutes, with real time labels along the x-axis, a clickable legend for hiding individual lines, and memory shown as used + percentage per container
- Diagnostics:
Remote Pulse: Show Logopens an output channel that records collector failures, availability-probe results and pause/resume events (each distinct error is logged once, not on every poll);Remote Pulse: Export Snapshot Historywrites the in-memory history plus the effective config to a JSON file in the system temp directory, which is what to attach when reporting an issue - Adaptive polling: automatically throttles once the window loses focus, reducing load on the remote machine
- Localized UI: commands, settings, and the status bar/webview text follow VS Code's display language (English by default, with a 简体中文 translation)
Search for Remote Pulse in the VS Code Extensions view, or install directly from the VS Code Marketplace or the Open VSX Registry (e.g. for VSCodium and other Open VSX-based editors).
Alternatively, download the .vsix file from Releases, then in VS Code run:
Extensions: Install from VSIX...
Or install from the command line:
code --install-extension remote-pulse-<version>.vsixOnce installed, connect to a Linux remote host over Remote-SSH and the metrics will show up in the status bar (the extension declares extensionKind: workspace, so it automatically runs on the remote extension host — no extra setup required).
| Setting | Default | Description |
|---|---|---|
remotePulse.enabled |
true |
Master switch. When off, polling stops, every status bar metric item is hidden, and the icon shows a paused state that resumes on click |
remotePulse.refreshInterval |
2000 |
Refresh interval for high-frequency foreground metrics (CPU/memory), in ms |
remotePulse.backgroundInterval |
15000 |
Throttled refresh interval once the window loses focus, in ms |
remotePulse.heavyMetricInterval |
10000 |
Independent polling interval for low-frequency metrics like GPU/Docker, in ms |
remotePulse.warningThreshold |
80 |
Warning threshold (%) |
remotePulse.criticalThreshold |
95 |
Critical threshold (%) |
remotePulse.statusBarMetrics |
["cpu", "memory"] |
Which metrics to show as status bar items — cpu, memory, disk (fullest mount point), gpu (primary GPU only), network (download/upload shown separately with arrow icons); unselected metrics still appear in the trend panel. Run Remote Pulse: Configure Status Bar Metrics for a real multi-select picker |
remotePulse.trendPanelSections |
["gpu", "docker"] |
Which optional sections/rows to show in the trend panel body (GPU cards, Docker table, the network row in "System"); System and Storage are always shown. Independent of trendChartMetrics — this doesn't affect the chart. Run Remote Pulse: Configure Trend Panel Sections for a real multi-select picker |
remotePulse.trendChartMetrics |
["cpu", "memory"] |
Which metrics to plot as lines in the trend panel's 30-minute chart — cpu, memory, gpu (primary GPU only), network (download and upload as two separate lines sharing their own right-hand axis). Independent of trendPanelSections and statusBarMetrics. Run Remote Pulse: Configure Trend Chart Metrics for a real multi-select picker |
remotePulse.enableNotifications |
false |
Whether to show a system notification when CPU, memory or disk crosses the critical threshold. The notification offers Show Trend and Mute for 10 minutes |
remotePulse.diskMountPoints |
[] |
Mount points to monitor; leave empty to show every real mount point, sorted by usage |
Remote Pulse: Show Trend Chart(remotePulse.showTrend, also bound to clicking the CPU/memory/disk/GPU/network status bar items)Remote Pulse: Refresh Now(remotePulse.refresh)Remote Pulse: Configure Status Bar Metrics(remotePulse.configureStatusBarMetrics, also bound to clicking the alert icon)Remote Pulse: Configure Trend Panel Sections(remotePulse.configureTrendPanelSections)Remote Pulse: Configure Trend Chart Metrics(remotePulse.configureTrendChartMetrics)Remote Pulse: Pause / Resume Monitoring(remotePulse.toggleEnabled, also bound to clicking the paused icon; resuming clears the history so the chart doesn't span the paused gap)Remote Pulse: Show Log(remotePulse.showLog)Remote Pulse: Export Snapshot History(remotePulse.exportSnapshot)
- Non-Linux remote hosts: CPU/memory automatically fall back to Node.js's
osmodule (slightly less precise); the network module is hidden entirely since there's no cross-platform equivalent - First connection: the status bar initially shows a
$(sync~spin)loading state - Collection failure (permissions / network flakiness): shows
$(circle-slash), with the failure reason in its tooltip and the full detail in the output channel — no intrusive error notifications - GPU/Docker unavailable: the section explains which case it is (not installed, no permission, or still waiting for the first sample) instead of disappearing. Availability is probed with a throttle rather than once at startup, so a daemon or driver that becomes ready later is still picked up
npm install
npm run lint # eslint (flat config, type-aware rules over src/)
npm run build # compile with tsc into out/
npm test # build, then run the unit tests under test/ (node:test)
npm run test:integration # runs test/integration/ in a real VS Code extension host (@vscode/test-cli)
npm run package # vsce package to produce a .vsixOpen this project in VS Code and press F5 to launch an Extension Development Host for live debugging (locally on macOS/Windows, CPU/memory fall back to the os module path, so you can verify the core interactions even without a remote Linux host).
The CI/release pipeline (GitHub Actions workflows, grabbing a PR's test build) and the one-time repository secrets setup needed for automatic publishing are documented in CONTRIBUTING.md.