skytells cloud deploy

Provision a cloud infrastructure instance — including NVIDIA GPU instances — with the Skytells CLI.

Overview

skytells cloud deploy provisions a new cloud infrastructure instance in the linked project. Instances can range from standard compute environments to high-performance GPU nodes including NVIDIA H100 and A100. This is the starting point for any cloud infrastructure provisioning workflow.

Syntax

skytells cloud deploy [--gpu <type>] [--region <region>] [--name <name>] [--json]

Flags

Flag	Description
`--gpu <type>`	GPU type to provision. See supported values below. Omit for CPU-only instances.
`--region <region>`	Target region for deployment.
`--name <name>`	Display name for the instance.
`--json`	Output the provisioned instance details as JSON.

GPU types

`--gpu` value	Hardware	Use case
`H100`	NVIDIA H100 80GB SXM	LLM training, large inference workloads
`A100`	NVIDIA A100 80GB	ML training, batch inference
(omitted)	CPU-only	Web services, APIs, workers

Examples

# Deploy a standard cloud instance
skytells cloud deploy --name "api-worker"

# Deploy an H100 GPU instance
skytells cloud deploy --gpu H100 --name "llm-trainer"

# Deploy an A100 instance in a specific region
skytells cloud deploy --gpu A100 --region us-west-2 --name "inference-server"

# Deploy and capture the instance ID
INSTANCE_ID=$(skytells cloud deploy --gpu H100 --json | jq -r '.id')
echo "Provisioned: $INSTANCE_ID"

Example output:

✔ Cloud instance provisioned

  ID:       inst_abc123
  Name:     llm-trainer
  GPU:      H100 (80GB SXM)
  Region:   us-east-1
  Status:   starting

After deploying

Once provisioned, check instance status:

skytells cloud inspect inst_abc123

List all instances:

skytells cloud ls

GPU instances take a few minutes to become running. Use skytells cloud inspect <id> --json | jq -r '.status' to poll for readiness, or wait for the Console to show an active state.

Billing

GPU instances accrue compute charges from the moment they reach running state. Always stop or destroy instances when not in use.

# Stop a GPU instance when not in use
skytells cloud halt inst_abc123

# Permanently destroy when no longer needed
skytells cloud destroy inst_abc123

A halted GPU instance still holds its GPU reservation and may accrue a smaller standby charge. Use skytells cloud destroy to fully release resources.

How is this guide?