DocsIntroduction

Documentation

Introduction

Kracht is the API and CLI for on-demand GPU compute. Launch a GPU in seconds, run your workload, and pay per second from your terminal, Python, or raw HTTP.

kracht launch --gpu h100api key readyjob.status runningdocs indexed

Quickstart

From setup to a sample H100 launch in three commands.

kracht · deploy
$

Pick your interface

Explore

Discover the full range of features and capabilities.

Provider onboarding

Lending a machine is one command. The agent runs unprivileged, isolates each rental, and reports capacity so the marketplace can match it.

Install the agentUbuntu · sudo
  1. 01

    Register the machine

    Add a host in your dashboard. You get an enrollment token, shown once.

  2. 02

    Run the command on it

    One line, on a fresh Ubuntu box. Idempotent, so re-running it upgrades the agent.

  3. 03

    It appears in your fleet

    The agent registers itself and starts reporting. Listed capacity, earning when rented.

curl -fsSL https://cp.kracht.ai/install.sh | sudo bash -s -- \
  --control-plane-url https://cp.kracht.ai \
  --enrollment-token <YOUR_TOKEN>

<YOUR_TOKEN>is issued when you register a machine, and shown once. It is bound to that machine from the moment it exists, so a leaked token cannot enroll someone else's box as a host.

What this script does to your machine
  • Installs prerequisites with apt, including Docker.
  • Creates an unprivileged system user, kracht-agent. Nothing runs as root after setup.
  • Installs the node agent as a systemd service so it survives a reboot.
  • Sets up an isolated session network, reapplied at every boot.
  • Optionally hardens SSH to key-only with no root login (--harden-ssh).
  • Optionally configures an nftables firewall, default-deny inbound (--firewall).
  • Optionally enables Docker user-namespace remapping, so a renter's container cannot map to root on your host.

Was this page helpful?

Two sides, one marketplace

Spin up the GPUs you need - or earn from the ones you already own

Rent on-demand compute by the second, or list idle hardware and let it pay for itself. Same marketplace, both directions.

No card required

Stay in the loop

Product updates and early access to new GPU classes. No spam.