> See the full index of documentation pages at https://docs.zettabyte.com/llms.txt.

# Quickstart

> Sign in, launch a GPU, deploy a model, and call it over the inference API.

Source: https://docs.zettabyte.com/docs/get-started/quickstart

Last updated: 2026-07-21

This walkthrough takes you from a new sign-in to your first call against a deployed inference endpoint.

## Before you begin

- An account on app.zettabytecloud.com (username and password).
- A bound payment method, unless your tenant is whitelisted for assigned resources. See Sign in and accounts and How billing works.

New to zCLOUD? Read How zCLOUD works first. It covers the mental model (the two workspaces, instances versus tasks, images, and storage) that the steps below assume.

## Steps

1. Sign in. Go to app.zettabytecloud.com and sign in. You land on the **Instance** page in the zCLOUD workspace.
2. Launch a GPU instance. On the **Instance** page, select **Launch Instance**. Set Resource Type to Standard GPU Instance, choose a Computing Config (for example `NVIDIA-A10-PCIE-4GB`), set the Count, and confirm the Purchase Cycle (On-Demand, billed hourly). Select **Confirm**, or **Join The Queue** if capacity is not immediately available.
3. Develop or run a job (optional). With a resource available you can open a Web IDE (`Jupyter` or `VSCode`), or create a training or inference task under **Tasks**. The deploy path below uses a pre-trained model and does not require this step.
4. Deploy a model. In Library > Models, choose a pre-trained model and select **Deploy** to open the inference task wizard. Set the Inference Engine (for example `vLLM`), the resource, the Number of Copies (replicas), storage, and optional autoscaling. The endpoint appears under Inference in zSTUDIO and under Tasks > Inference in zCLOUD. It starts as Pending and is ready once its replicas are running.
5. Create an API key. Open **API KEY** and select **Create API KEY**. Give it a unique name, grant the Inference Service permission, and set a validity period. The key is **shown once** at creation; copy it and store it securely.
6. Call your endpoint. Once the endpoint is running, call it over the inference API. The `Service-Name` header binds the request to your deployment, using your tenant ID and the deployment's job alias name.

> [!NOTE]
> **After you launch the instance (step 2)**
> Billing starts as soon as the instance is allocated, not when you start using it. If capacity is not available, select **Join The Queue**. See The resource queue.

> [!NOTE]
> **After you deploy the model (step 4)**
> A new endpoint starts as Pending and is ready once its replicas are running (Scaling shows, for example, `1/1`).

```bash
curl https://app.zettabytecloud.com/inference/v1/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Service-Name: TENANT_ID.JOB_ALIAS_NAME" \
  -d '{
    "prompt": "Hello",
    "max_tokens": 128,
    "temperature": 0.7
  }'
```

See Inference API for the full request and response reference.

## Troubleshooting

- Endpoint stuck on Pending: the replicas are still scheduling or capacity is short. See The resource queue.
- `401` on the API call: your key is missing the Inference Service permission or has expired. See API keys.
- Wrong or empty response: check the `Service-Name` header format (`TENANT_ID.JOB_ALIAS_NAME`). See Call your endpoint.

## Next steps

- Inference API: full request and response reference.
- How billing works: wallet, payment, and consumption.