Skip to content
runcrate
Console

CLI

GPUs from a shell. Run by your agent.

For Claude Code, Codex and any agent that runs commands. Launch a GPU, SSH in, copy files and delete it, from $0.20/hr.

Install the CLIRead the CLI docs

macOS, Linux and Windows. One binary.

Terminal
The command: runcrate instances types --gpu A6000
Outputstdout
exit 00.00 s

Recorded from runcrate v0.2.0 on October 19, 2026. The address and ids are hidden.

Install

One line, then sign in. Your browser opens. No key to paste.

  1. 01

    Install it.

    One line on macOS or Linux, Homebrew, or PowerShell on Windows. One binary, nothing to install first.

  2. 02

    Sign in.

    runcrate login opens your browser to sign in and pick a workspace. Approve it and the CLI stays signed in.

  3. 03

    Tell your agent.

    Claude Code and Codex run commands in your terminal, so they can use it as soon as it is installed. Ask for what you want and say to use the runcrate CLI.

terminalbash
curl -sSL https://runcrate.ai/install.sh | sh
export PATH="$HOME/.local/bin:$PATH"

macOS and Linux. The second line puts it on your PATH in this shell.

On a server or in CI, nothing opens a browser: set RUNCRATE_API_KEY to a key from the console and the CLI uses it instead.

Commands

Everything a GPU job needs. From the first price to the last delete.

  • runcrate instances

    See every GPU on offer with its hourly price, launch one, check on it, delete it.

    types · create · status · delete

  • runcrate ssh

    A shell on a running instance, or one command and out. Each session gets a short-lived certificate, so there is no key to manage.

    ssh my-gpu -- nvidia-smi

  • runcrate cp

    Copy files and whole folders to an instance and back.

    -r for whole folders

  • runcrate volumes

    Storage that outlives the instance: create it, resize it, delete it. Attaching one at launch is in the SDK and the API.

    create · resize · delete

  • runcrate templates

    The crates an instance can boot: Devbox, Finetune and Inference, with what each has installed.

    3 crates

  • runcrate billing

    What is left on your balance, and what your model calls spent. Instance charges are in the console.

    balance · usage

  • runcrate envs

    Keep staging apart from production inside one workspace, and switch between workspaces.

    envs · workspaces

  • runcrate ssh-keys

    Add your own public key, for tools that connect without the CLI.

    list · add · delete

The CLI runs GPUs, storage and billing. Model calls go through the API or the SDKs, with the same key and the same balance.

Every command in the docs

Output

Output a script can trust. Data on stdout, errors on stderr.

  • --json

    Machine-readable output on list commands, for a script or an agent to parse.

    most list commands

  • -q

    Only the ids, one a line, ready to pipe into the next command.

    instances list

  • exit 1

    Any failure exits 1 with the reason on stderr and nothing on stdout. Success exits 0.

    $?

  • RUNCRATE_API_KEY

    Set a key from the console and the CLI uses it instead of a browser sign-in. Right for CI and cron.

    environment

runcrate instances create returns as soon as the instance is queued; poll runcrate instances status until it reads running, as the session above does.

Examples

Ask, and your agent runs it. The commands it reaches for.

Claude Code or Codex turns one request into the commands: launch, copy, run, delete.

Read the instance commands
Start an H100 with the Finetune crate, copy train.py to it, run it, and delete the instance when it finishes. Runs runcrate instances create --name my-gpu --gpu H100 --template ubuntu-train. Runs runcrate instances status my-gpu. Runs runcrate cp ./train.py my-gpu:/workspace/. Runs runcrate ssh my-gpu -- python /workspace/train.py. Runs runcrate instances delete my-gpu.

Pricing

Same GPUs, same rates. Billed for the time they run.

GPU

From, per hour

  • H100 SXM

    80GB of memory

    $3.25/hr

  • A100 80GB

    80GB of memory

    $1.76/hr

  • L40S

    48GB of memory

    $1.14/hr

  • RTX 4090

    24GB of memory

    $0.52/hr

  • An instance bills from your dollar balance while it runs: by the second on most GPUs; a few bill in whole minutes, rounded up, including about 30 seconds of teardown after you delete.

  • runcrate instances types prints every offer's hourly price before anything launches.

  • runcrate billing balance shows what is left; runcrate billing usage shows what model calls spent. Instance charges are under Billing in the console.

See every GPU price

FAQ

Questions about the CLI.