Getting started

Create an account, add credits, and deploy your first pod — usually under two minutes end to end. This guide walks through the full first-time setup.

Prerequisites

You need a verified email, an SSH key pair, and a payment method on file before deploying production workloads.

  1. Sign up at /register and verify your email.
  2. Add an SSH public key in Settings — it will be installed automatically on every pod you deploy.
  3. Add credits via the Billing page. Credits never expire and apply to any product.
  4. Open Pods, pick a template and GPU tier, then click Deploy.
  5. Wait 30–90 seconds for the pod to enter the running state.
# Example: verify your setup locally
ssh -i ~/.ssh/id_ed25519 root@<pod-ip>

Once connected, you have full root access to a container with NVIDIA drivers, CUDA, and common ML runtimes preinstalled. From there, attach Jupyter, mount volumes, or run training jobs.

Connecting

Every running pod exposes three connection options: SSH, browser-based web terminal, and Jupyter Lab. Use whichever fits your workflow.

SSH

Click Connect on the Pods page to copy an SSH command. Your key file is injected at deploy time from Settings — make sure it is configured before creating the pod.

ssh -o StrictHostKeyChecking=accept-new -i ~/.ssh/id_ed25519 root@<pod-ip>
Common mistake

If SSH asks for a password, your local agent isn't offering the right key. Explicitly path it with -i, or add the key to your ssh-agent.

Web terminal

No local SSH setup is required. The in-browser terminal opens directly to your pod shell over WebSocket and behaves like a normal terminal.

Jupyter

Most templates ship with Jupyter Lab preinstalled. Click the notebook icon in the Connect dropdown. Jupyter runs over HTTPS and uses your Spin Up GPU login.

Storage

Network volumes are persistent disks that exist independently of any pod. Data survives pod restarts, resizes, and replacements.

  • Create a volume by selecting a datacenter and size on the Storage page.
  • Attach one or more volumes at deploy or while the pod is running.
  • A pod's container disk is ephemeral — never store models or datasets on it.
  • Volumes bill separately from compute based on GB/month and duration.
Performance note

For dataloader-heavy training, prefer instances with local NVMe. Network volume throughput varies by datacenter.

Serverless

Serverless endpoints scale from zero to many workers automatically, and bill per second of actual usage. Ideal for batched inference, APIs, and async endpoints.

Deploy options

  • Hub template — one-click models such as Llama, Mixtral, SDXL.
  • Docker image — bring your own serving stack.
  • Hugging Face model via vLLM, TGI, or Text Generation Inference.
  • GitHub repo — build from source on first spin.

Scaling knobs

  • Min workers: 0 keeps cost at zero while idle; 1+ keeps a worker warm.
  • Max workers: your hard ceiling.
  • Idle timeout: seconds of no requests before scaling down from min workers toward zero.
Cold starts

With min workers at 0, first-request latency can spike during scale-out. For latency-sensitive traffic, set min workers to 1.

Billing & Usage

Credits are consumed per second while pods run and while serverless workers are active. Nothing is charged while resources are stopped.

  • Top up any amount from the Billing page via Stripe.
  • Your balance is always visible in the console header.
  • Volumes and outbound traffic bill at separate rates shown on the Pricing page.
  • If balance reaches zero, running pods may be automatically stopped.
Teams

Organization billing lets you set shared limits, view spend per project, and issue API keys scoped to teammates.

Troubleshooting

"There are no instances currently available"

The selected GPU/CPU tier is temporarily out of stock in that datacenter. Switch to a nearby region or a different SKU, then retry.

SSH password prompt instead of key login

Confirm the key exists in Settings before the pod starts. Keys are not injected retroactively into already-running pods.

# Debug key auth
ssh -v -i ~/.ssh/id_ed25519 root@<pod-ip>

Pod stuck starting

Most pods are ready in under 90 seconds. First-time template pulls or large custom images can take several minutes longer. If a pod stays past 10 minutes, stop and recreate it — that usually resolves the state.

Still stuck?

Open a ticket via Contact Sales and include the pod ID.

← Products Pricing →