Request capacity

Quickstart

From a new account to a running, attested node in five steps.

Updated 15 October 2026

This guide takes you from an empty project to an eight-GPU node that you have verified before use. It assumes you have console access; during the private preview, access is granted per organisation.

1. Install the CLI

The CLI is a single static binary for Linux and macOS.

curl -fsSL https://cloisterlabs.com/install.sh | sh
cloister version

Authenticate with a token created in the console under Project → Access tokens. Tokens are scoped to one project and can be limited to read-only operations.

export CLOISTER_TOKEN=clt_live_…
cloister auth status

2. Choose a plan

Every configuration has a plan identifier that names the accelerator and the GPU count. List what you can provision in a region:

cloister plans list --region eu-n

PLAN            GPUS  MEMORY/GPU  VCPU  RAM      NVME     AVAILABILITY
cl.h200.x8      8     141 GB      224   2 TB     30.7 TB  available
cl.h100.x8      8     80 GB       224   2 TB     30.7 TB  available
cl.l40s.x4      4     48 GB       64    768 GB   7.7 TB   available

Regions, availability and prices shown in this documentation are sample data until the inventory service is connected.

3. Create a node

Pass --confidential to boot with measured firmware and, on supported accelerators, GPU confidential-computing mode enabled. See Confidential mode for what that changes.

cloister nodes create \
  --plan cl.h200.x8 \
  --region eu-n \
  --ssh-key ~/.ssh/id_ed25519.pub \
  --confidential

The node moves through provisioning → measuring → ready. A node is only assigned to you after its drives have been erased and its firmware re-flashed from signed images.

4. Verify before you use it

Fetch the attestation evidence and check it against a policy you control. Nothing here relies on trusting Cloister Labs: the evidence is signed by keys rooted in the hardware vendors’ certificate chains.

cloister attest verify node_7Hq2 --policy ./policy.rego

The Attestation guide explains the evidence format and how to write a policy.

5. Connect

ssh ubuntu@$(cloister nodes get node_7Hq2 --field ipv4)
nvidia-smi

When you are finished, delete the node. Billing stops at the second the delete is accepted.

cloister nodes delete node_7Hq2

Next steps

  • Read how nodes are built and what you can change on them.
  • Release disk or model keys only to verified nodes with the Attestation guide.
  • Automate the same steps with the API.