vitals · v0.10.0 · macOS · Linux · Windows

Names the one thing slowing your machine down — and the exact command to fix it.

A single static binary that correlates CPU, memory, disk, network, GPU and local/cloud LLM signals into one ranked verdict, instead of a wall of gauges you have to interpret yourself.

Install vitals View on GitHub Read the user guide
zsh — vitals doctor
$ vitals doctor
✗ Swap thrashing
swap 91% full and paging out at 40.00 MB/s — the machine is stalling on disk,
not working — largest consumer: Chrome Helper (Renderer) (pid 4821), 1.20 GB RSS
→ quit or restart Chrome Helper (Renderer) (pid 4821) — the largest consumer, or run vitals memhogs for the full list
→ macOS: sudo purge
✗ verdict: CRITICAL — address the finding above now
exit 2
What it does

One binary. Every resource. One verdict.

vitals reads the same system data htop/btop, ncdu and nvtop already read — it doesn't replace them. What it adds is the layer above: correlating CPU, memory, swap, disk, network, power and GPU into a single ranked list of what's actually wrong, with a concrete fix for each finding, not just a number.

Cross-resource correlation

High CPU that's really disk I/O wait. Swap thrashing vs. reclaimable cache. A single-threaded bottleneck vs. one that needs more cores.

vitals doctor

Local & cloud LLM insight

Per-model GPU VRAM offload percentage for Ollama, LM Studio, llama.cpp and vLLM — plus reachability across 11 open-API cloud providers, only probed when you've set the key.

vitals llm · vitals llm fit <model>

The whole machine on one screen

A verdict-first, single-screen terminal view: the ranked finding up top, then tiled CPU / memory / disk / network / power / GPU panels and the heaviest processes. Bare vitals on a terminal, or vitals view.

vitals view

Apply the fix, with confirmation

Where a finding has a safe, reversible remedy, vitals heal runs it for you — always previewed, always confirmed, from a compile-time allow-list. Everything else stays a printed instruction.

vitals heal --dry-run

A browsable dashboard

The same verdict as loopback-only web pages that refresh themselves in place: a grouped sidebar, a page per resource with the top consumers on each — biggest files, active connections, processes holding VRAM, per-process disk I/O — plus a Containers page, LLM offload and machine details. Opens your browser automatically; nothing leaves your machine.

vitals dashboard

Local containers & Kubernetes

vitals probes a Docker-compatible runtime and a loopback Kubernetes context in one pass, read-only, showing each as its own section — OOM-kills, CrashLoopBackOff, restart loops, published ports, and memory the host process list can't see — with a docker/kubectl command for each.

vitals containers

Safe disk cleanup

Cross-platform cache/log/temp cleanup and byte-identical duplicate detection — always --dry-run first, never deletes without confirmation.

vitals clean --dry-run · vitals dupes

Scriptable by design

Exit codes (0/1/2) for CI and cron, --json with a versioned schema, a Prometheus exporter, and an MCP server for AI coding assistants.

vitals doctor --json | jq

No phone-home, minimal dependencies

gopsutil for system data, plus a small terminal-reliability group for cross-platform color and width detection. No telemetry, no accounts, no network calls unless you point it at an LLM endpoint yourself.

go build -o vitals .
See it

The same verdict, as a local web app

vitals dashboard serves the doctor's correlation as browsable pages — one per resource the machine actually has — on 127.0.0.1 only. Read-only pages refresh themselves in place; nothing leaves your machine.

vitals dashboard Overview page: a swap-pressure verdict banner, resource cards for CPU, memory, disk, network, power and GPU with trend sparklines, quick actions, and the ranked findings below.
Overview — the verdict, every resource at a glance with 24h trend, and the ranked findings.
The same vitals dashboard Overview page rendered in dark mode.
Follows your system light/dark preference. A grouped, icon-led sidebar reaches every resource page, plus Processes, Containers, LLM Insight and System. The Containers card sums running / unhealthy across Docker and a local Kubernetes context.

Real screenshots of vitals dashboard at v0.10.0 — regenerated when the layout changes (scripts/site-screenshots.mjs).

How it fits alongside what you already use

Complements, not a replacement

Each of these tools does one job well. vitals doesn't try to out-gauge them — it adds the correlation and remediation layer none of them provide on their own.

ToolWhat it's forWhat vitals adds
htop / btop Live per-process CPU, memory and load, interactively. Explains why a number is high (I/O wait vs. real compute) and what to do about it.
ncdu / gdu / dust Interactive disk-usage browsing, directory by directory. A one-line disk verdict correlated with the rest of the machine, plus safe automated cleanup and duplicate detection.
nvtop Live GPU utilization and VRAM per process. Per-model LLM offload percentage specifically, and whether a model is silently running CPU-only.
glances A single-screen overview of most system metrics at once. vitals view gives the same one-screen layout but leads with a ranked verdict, not a grid of numbers — plus JSON/Prometheus output and exit codes for automation.

This table describes each tool's own core purpose as of vitals v0.10.0 — check each project's own docs for its current full feature set.

Install

Get started

macOS / Linux

brew install gautampachnanda101/tap/vitals
# or, without Homebrew:
curl -fsSL https://raw.githubusercontent.com/gautampachnanda101/vitals/main/install.sh | sh

Windows

scoop bucket add gautampachnanda101 https://github.com/gautampachnanda101/scoop-bucket
scoop install vitals

Direct download

Grab a tarball from the latest release — prebuilt for linux/darwin/windows, amd64 and arm64.

From source

go build -o vitals .   # Go 1.27+