QuickPods

Deploy production-ready AI stacks in one click. QuickPods is Podstack’s managed, click-to-deploy product for GPU workloads. Pick a 1-click template, choose a GPU (full or fractional), and launch a running pod in seconds — no Dockerfiles, no Kubernetes, no infrastructure to wire up.

Every pod comes with persistent storage, SSH and a browser terminal, live logs and metrics, and a built-in MLOps stack for experiment tracking, model versioning, monitoring, and pipelines.

What is a QuickPod?

A QuickPod is a containerized GPU workload — a pod — running on Podstack’s managed infrastructure. Each pod runs a container built from a 1-click template (or any image you choose) with:

• GPU access — a full GPU or a fractional slice, across A100, H100, H200, L40S, V100, and T4 hardware
• CPU, memory, and disk allocation you control
• Persistent storage via NFS/data volumes that survive restarts and recreations
• Access through SSH, an in-browser web terminal, Jupyter, and auto-generated HTTPS URLs for any exposed port
• Live observability — streaming logs and real-time CPU/GPU/memory metrics

When to use QuickPods

QuickPods is the fastest path when you want to:

• Spin up a Jupyter or JupyterLab GPU notebook for experimentation
• Run a ready-made app — ComfyUI, Automatic1111, Ollama, vLLM, text-generation-webui — without building an image
• Fine-tune or train a model (Axolotl, Unsloth, LLaMA-Factory, Kohya)
• Serve an LLM or embedding model on a GPU
• Track experiments, register models, and monitor them from one place

If you need dedicated, non-virtualized hardware for large-scale training, see the GPU Marketplace instead. If you need auto-scaling, OpenAI-compatible serving endpoints, see Inference.

Capabilities

What’s available today

These features are live in the portal right now — launch a pod and you can use all of them:

• Pods list and pod detail — see every pod, its status, and drill into one for logs, metrics, and connection details.
• One-click launch — a three-step GPU → template → configure flow, with a join-the-waitlist option when a GPU type is temporarily out of capacity.
• 1-click template catalog — the pre-built app catalog plus any custom templates you’ve saved; deep-link into launch from the Templates page.
• Fractional GPUs — request a whole card or a percentage slice of one when you configure a pod. See Manage & Scale.
• Persistent volumes — create and attach volumes at a mount path during launch, and manage them from a standalone Volumes page. See Storage & Data.
• Live logs and real-time stats — streaming container logs and CPU/GPU/memory metrics on the pod detail page.
• In-browser web terminal — a full terminal for any running pod, no local setup.
• SSH access — per-pod SSH connection details plus an SSH Keys page to register your keys. See Launch a Pod.
• Save your own templates — turn a working pod configuration into a reusable 1-click template for your team.
• Wallet and invoices — per-second billing, wallet balance, and invoices, all in the portal.

Not yet generally available: the built-in MLOps suite (experiment tracking, model registry, monitoring, drift detection, pipelines, and schedules) is built but currently gated behind a feature flag, so it may not appear in your portal. If the MLOps/registry sections aren’t visible, they aren’t enabled for your account yet — see MLOps or contact support.

Use cases

• Notebook experimentation — a researcher launches a JupyterLab template on an L40S, attaches a volume at /data, iterates in the notebook, then stops the pod to pause billing.
• Run a ready-made app — a creative technologist launches the ComfyUI (or Automatic1111) template and opens its auto-generated HTTPS URL to start generating images, with no image to build.
• Serve a model for a prototype — a developer launches the Ollama or vLLM template on a fractional GPU, exposes the port, and hits an OpenAI-style endpoint from their app.
• Fine-tune in a prebuilt environment — an ML engineer picks an Axolotl or Unsloth template, opens the web terminal, and runs a LoRA job against a dataset that lives on a persistent volume.
• Cost-controlled dev box — a student rents a fractional slice of a GPU for light development, using the web terminal and live metrics, and pays only for the seconds it runs.
• Standardize a team environment — a team lead configures a pod once (image, ports, env vars), saves it as a custom template, and the whole team launches identical environments in one click.

The path

Work through these pages in order, or jump to the one you need:

1. Launch a Pod — the full click-by-click launch flow.
2. Templates — the 1-click template catalog and how to use and save templates.
3. Storage & Data — volumes, the /data convention, and moving files in and out.
4. Manage & Scale — start/stop, logs, metrics, resizing, and fractional GPUs.
5. MLOps — experiment tracking, model registry, monitoring, pipelines, and schedules.
6. Scenarios & Walkthroughs — end-to-end examples you can follow along with.
7. Troubleshooting — fixes for the most common issues.

FAQs

What’s the difference between a QuickPod and a “pod”? They’re the same resource. “QuickPod” is the managed, template-first launch experience in the portal; the underlying resource is a pod (container) you manage on the pod detail page.

Do I need to know Docker or Kubernetes? No. Pick a 1-click template and launch. Everything — the image, ports, and environment — is pre-configured. Advanced users can still supply a custom image and settings.

How is billing calculated? Per second, based on the GPU fraction, CPU, memory, and disk you allocate. You only pay while a pod is Running. Stop the pod to pause billing while keeping its configuration.

What happens to my data when I stop or delete a pod? Data written inside the container (ephemeral disk) may be lost on stop or delete. Anything on an attached volume — mounted at /data by default — persists. Always keep important files under /data. See Storage & Data.

Can I use a fractional GPU? Yes. When you pick a GPU you can request a slice rather than a whole card, which lowers the hourly rate for light inference and development. See Manage & Scale.

How do I connect to my pod? Over SSH, through the in-browser web terminal, via Jupyter (on notebook images), or through an auto-generated HTTPS URL for any port you expose. See Launch a Pod.

My GPU type shows as unavailable — what now? Choose a different GPU type, reduce the count, or join the waitlist to be notified when capacity frees up. See Troubleshooting.

Is MLOps available on every account? MLOps features may be gated by a feature flag or account level. If the MLOps/registry sections aren’t visible, contact support.