GPU CLI

Your GPU cloud,
one command away

The gpu CLI launches the cheapest GPU on the market, wires up SSH, and streams your logs, all without a single dashboard.

$curl -fsSL https://gpu.ai/install | sh

Get started Read the docs

zsh · gpulive

Platform

One control plane for every GPU

The CLI and dashboard share one engine: launch, watch spend, and track capacity across regions in real time.

gpu control planeus-east

Instances2 running

gpu-7f3a

4× H100 · us-east

$10.56/hr

running

gpu-2b91

1× A100 · eu-west

$2.00/hr

running

gpu-c4d7

8× H200 · ap-south

$30.48/hr

provisioning

gpu-9e10

1× L40S · us-west

$0.00/hr

idle

Live spend

$12.56/hr

across 2 instances

$128.40

billed this session

price scanslive · 30s

Network

us-west

us-central

us-east

ca-central

eu-west

eu-east

india-west

india-south

asia-pacific

Commands

A verb for everything you do

gpu instances create

Provision the cheapest matching GPU and poll until it's running.

$ gpu instances create -t h100_sxm

gpu instances ssh

Resolve the FRP tunnel and drop into the instance over SSH.

$ gpu instances ssh gpu-7f3a1c

gpu instances list

List every instance in your org with status, GPU, and count.

$ gpu instances list

gpu pricing

Show the live USD/hour price grid and availability by GPU type.

$ gpu pricing --gpu-type h100_sxm

gpu chat

Run a serverless chat completion that streams tokens to stdout.

$ gpu chat gpuai/qwen2.5-7b-instruct "hi"

Why the CLI

Built for builders

One binary, zero config

Install once and authenticate. No SDKs to wire up, no cloud consoles, no Terraform.

Scriptable & CI-ready

Output pipes straight into jq. Drop gpu into any pipeline or Makefile.

SSH that just works

Keys, tunnels, and host config are provisioned automatically. No firewall dance.

Per-second metering

Spend is tracked live and stops the moment you do. No minimums, no hourly rounding.

Runs onmacOSLinuxWindowsany CI runner

From zero to training in one command