gpunz — command-line client for the gpu.nz GPU cloud

Rent an RTX 4090, A100 or H100 from your terminal, billed by the minute. Search live GPU offers, launch ComfyUI, Jupyter, Ollama or vLLM in one command, SSH in, stop when you are done.

Website · Pricing · CLI docs · REST API · Examples · Русский

---

Why gpunz

• Per-minute billing. A stopped server stops costing GPU time; you see the hourly price before you rent.
• Any GPU. From RTX 3060 to RTX 4090/5090, A100, H100, H200 and B200, on demand or interruptible (cheaper).
• One-command environments. Templates for ComfyUI, Stable Diffusion Forge, JupyterLab, Ollama + Open WebUI, vLLM (OpenAI-compatible API), PyTorch, or your own Docker image.
• Scriptable. Every command has --json, so it fits shell scripts, CI and agents.
• Top up from the terminal. gpunz topup 500 draws a QR code you scan with your phone to pay by SBP.
• Single static binary. Linux, macOS and Windows, amd64 and arm64. No runtime to install.

Install

# Linux / macOS
curl -fsSL https://gpu.nz/install.sh | sh

# Windows (PowerShell)
irm https://gpu.nz/install.ps1 | iex

# Homebrew (macOS, Linux)
brew install gpunz/tap/gpunz

# Scoop (Windows)
scoop bucket add gpunz https://github.com/gpunz/scoop-bucket
scoop install gpunz

# From source (Go 1.22+)
go install github.com/gpunz/gpunz-cli/cmd/gpunz@latest

The installers verify the binary against SHA256SUMS before installing; Homebrew and Scoop check the same hashes. Using an AI agent? The gpu.nz MCP server gives Claude, Cursor and Codex the same actions. Prebuilt binaries are also attached to every GitHub release. Update later with gpunz update.

Quick start

gpunz login                                   # opens the browser to approve this device
gpunz topup 500                               # pay by SBP: scan the QR code in the terminal with your phone
gpunz gpus                                    # GPU models, VRAM, free units and the lowest price
gpunz search --gpu rtx-4090-24gb --limit 5    # live offers, cheapest first
gpunz templates                               # ComfyUI, Jupyter, Ollama, vLLM, ...
gpunz create OFFER_ID --template comfyui --disk 60
gpunz instances                               # your servers and their status
gpunz ssh INSTANCE_ID                         # SSH straight into the server
gpunz stop INSTANCE_ID                        # stop paying for the GPU, keep the disk
gpunz destroy INSTANCE_ID                     # delete the server and its disk

Example output:

$ gpunz search --gpu rtx-4090-24gb --limit 3
ID          GPU          VRAM   CPU     RAM     DISK     NET ↓/↑      RELIAB.  COUNTRY  ON DEMAND   INTERRUPTIBLE
50hfa8895b  RTX 4090 ×1  24 GB  —       7 GB    1440 GB  347/10 Mbps  96.3%    US       32.90 ₽/h   —
nma699m6rx  RTX 4090 ×1  24 GB  —       7 GB    1440 GB  658/18 Mbps  96.5%    US       32.90 ₽/h   —
hh5qwswn7b  RTX 4090 ×1  24 GB  8 vCPU  126 GB  329 GB   840/98 Mbps  99.5%    AU       60.90 ₽/h   39.90 ₽/h

Prices change with the market; the CLI always shows the live price.

Commands

Command	What it does
login [--key KEY] [--no-browser]	Sign in with the browser device flow, or save an API key
logout / whoami	Forget the saved key / show the signed-in account
balance	Balance and how long it lasts at the current spend
topup [AMOUNT]	Top up: opens the payment page (SBP or crypto) and draws its link as a QR code for your phone camera
gpus	GPU models with VRAM, availability and the lowest price
search [filters]	Rentable offers: --gpu, --num-gpus, --type on_demand|interruptible, --max-price, --country, --min-vram, --min-cuda, --min-disk, --min-down, --verified, --order, --limit
templates	Launch templates (ComfyUI, Forge, Jupyter, Ollama, vLLM, PyTorch, ...)
create OFFER_ID --template SLUG	Rent an offer: --disk GB, --name, --ssh-key, --env KEY=VALUE, --onstart
instances / ls	Your servers (--all includes deleted ones)
show ID / logs ID [--tail N]	Details and logs of one server
start ID / stop ID / destroy ID [-y]	Lifecycle; a stopped server keeps its disk
ssh ID [ssh args...]	Open an SSH session
ssh-keys [ls|add PATH|rm ID]	Manage SSH keys
api-keys [ls|create --name N|rm ID]	Manage API keys for scripts and CI
update, version, completion bash|zsh|fish|powershell	Self-update, version, shell completion

Global flags: --json (machine-readable output), --api-key KEY, --api-url URL.

Scripting

# cheapest on-demand RTX 4090 as JSON
gpunz --json search --gpu rtx-4090-24gb --type on_demand --limit 1 | jq '.data[0]'

# use an API key in CI instead of the browser login
export GPUNZ_API_KEY=gpz_xxx
gpunz --json instances

Configuration lives in your user config directory; GPUNZ_API_KEY and GPUNZ_API_URL override it. More recipes, Python and notebooks: gpunz/examples.

Build from source

go test ./...
go build -o gpunz ./cmd/gpunz
./build.sh   # cross-compile every platform into dist/ with SHA256SUMS

Security

The CLI talks only to https://gpu.nz/api/v1 over HTTPS, stores the API key in your user config file with owner-only permissions, and verifies downloaded updates against SHA256SUMS. Please report vulnerabilities privately: see SECURITY.md.

---

Русский

gpunz — консольный клиент облака gpu.nz: аренда видеокарт для нейросетей прямо из терминала, оплата поминутно, в рублях (СБП) или криптовалютой.

• Любая видеокарта: RTX 3060, RTX 4090/5090, A100, H100, H200, B200. По требованию или прерываемые, они дешевле.
• Готовые окружения одной командой: ComfyUI, Stable Diffusion Forge, JupyterLab, Ollama + Open WebUI, vLLM с OpenAI-совместимым API, PyTorch или свой Docker-образ.
• Поминутная оплата: цена за час видна до запуска, остановленный сервер не платит за GPU.
• Для скриптов: у каждой команды есть --json.

curl -fsSL https://gpu.nz/install.sh | sh     # Linux / macOS
irm https://gpu.nz/install.ps1 | iex          # Windows

gpunz login
gpunz topup 500                               # QR-код для оплаты по СБП с телефона
gpunz search --gpu rtx-4090-24gb --limit 5
gpunz create OFFER_ID --template comfyui
gpunz ssh INSTANCE_ID
gpunz stop INSTANCE_ID

Документация: gpu.nz/docs/cli · Цены: gpu.nz/pricing · Примеры: gpunz/examples

License

MIT © 2026 gpu.nz