Manage & Scale

Once a pod is launched, you manage it from the Pods list and each pod’s detail page.

The Pods list

Go to Pods to see every pod in the current project. Filter by project with the All Projects dropdown. Each row shows:

• Pod — name and container image
• Resources — vCPU, RAM, and GPU allocation (shown as 1x A100 for whole GPUs or 50% for a fractional slice)
• Status — the pod’s state (see below)
• Created — when it was launched
• Actions — the controls for that pod

Pod states

Start, stop, and delete

From the list row or the pod’s detail page header:

• Stop — appears on a running pod. Confirm in the Stop Pod dialog. Billing pauses, the GPU is released, and non-persistent data may be lost. The pod’s configuration is kept so you can start it again.
• Start — appears on a stopped pod. Billing resumes when it reaches running.
• Relaunch — appears on a failed pod. If the pod failed because it ran out of memory, the relaunch dialog lets you bump its memory before trying again.
• Delete — permanently removes the pod. Confirm in the Delete Pod dialog. There’s no pro-rated refund — you’re billed for the full billing cycle even if you delete early.

Save money: stop pods you aren’t actively using. Stopping pauses billing immediately while preserving the pod’s setup. Delete pods you’re finished with. Keep anything important on a volume so a stop or delete doesn’t lose it.

Low balance auto-stop. If your wallet balance runs out, running pods may be stopped automatically after a grace period. Keep enough balance (or enable auto-debit) to avoid interruption.

The pod detail page

Click a pod to open its detail page, which gathers everything about it into one scrolling view.

Web Terminal

On a running pod, click Open Terminal in the Web Terminal card for an interactive browser shell — no SSH client or keys required.

SSH Access

If you attached an SSH key, the SSH Access card shows a copy-ready command. Click Configure SSH for one-time setup instructions per OS. See SSH & PowerShell.

Notebook Access

On notebook images, the Notebook Access card shows the notebook URL, username (podstack), and an auto-generated password. Copy the password before leaving — it isn’t shown again.

Logs and metrics

• Container Logs stream live as your workload produces them.
• Infrastructure Stats show real-time CPU, memory, storage, and GPU utilization for the pod.

Billing

The Billing card shows the billing period, whether billing is Active/Paused/Not Started, the total billed so far, and the hourly rate breakdown across CPU, memory, storage, and GPU.

Exposed ports

Open a service to the internet after launch:

1. On the pod detail page, find Exposed Ports & Endpoints and click Edit.
2. In the Edit Exposed Ports dialog, enter a container port and click Add. Remove a port with the X next to it. (Port 22 can’t be removed while SSH is enabled.)
3. Click Save Changes — the change is applied to the running pod.

Each exposed port gets a public HTTPS endpoint:

• HTTP services: https://<subdomain>-<port>.cloud.podstack.ai
• SSH (port 22): ssh-<subdomain>.cloud.podstack.ai

Fractional GPUs

You don’t have to rent a whole GPU. When you configure a pod you set two independent controls (see Launch a Pod):

• Number of GPUs — how many GPUs to attach (1 or more).
• GPU Memory/Cores per GPU — the percentage of each GPU to use, from 5% to 100% in 5% steps. 100% is a whole card.

Picking a fraction lowers your hourly rate, which is ideal for light inference, development, and notebook work. The launch wizard’s donut chart shows how much of the GPU is already booked, how much you’re requesting, and how much is free, plus the effective VRAM. Fractional pods display their allocation as a percentage (for example 25%) in the list and on the detail page.

If your requested fraction exceeds what’s free right now, the pod may queue until capacity opens up.

Resizing a pod

CPU, memory, GPU count, and GPU fraction are set at launch and aren’t changed in place. To run with different compute:

• Change compute: launch a new pod with the new sizing (save your config as a template first to make this one click), point it at the same volume so your data carries over, then delete the old pod.
• Add memory to a failed pod: if a pod failed from running out of memory, use Relaunch to bump its memory.
• Change ports: edit exposed ports on a running pod any time (above).

Next steps

• Storage & Data — keep data across relaunches.
• Troubleshooting — fixes for common issues.
• Scenarios & Walkthroughs — end-to-end examples.