QuickPods
Go from zero to a running AI stack in minutes — pick a template, get MLOps out of the box, and run on fractional GPUs so you only pay for the slice you need.
Overview
QuickPods packages the tools most AI teams stitch together by hand — serving, fine-tuning, notebooks, experiment tracking, a model registry — into one-click templates on GPUs virtualized with our proprietary podvirt layer. No Dockerfiles, no cluster setup, no idle whole-GPU bills.
Loading live data…
Why teams choose QuickPods
Built on PodStack's proprietary PodVirt layer — not vanilla Kubernetes or OpenStack — so you get fractional GPUs, sub-second scaling, and one bill.
1-click templates for vLLM, PyTorch, TensorFlow, ComfyUI and Unsloth
Built-in MLOps: experiment tracking, model registry and metrics
Fractional NVIDIA GPUs from 12.5% to 100% via PodVirt
Per-minute billing with zero egress fees
Instant provisioning on operator-owned, ISO 27001 hardware
Bring your own Docker image or start from a template
From notebook to production in one click
QuickPods come with experiment tracking, a model registry, and real-time cost metrics baked in. Launch a ComfyUI image-generation environment, a vLLM inference server, or an Axolotl fine-tuning pod — all from the same console.
- ✓One-click ComfyUI, vLLM, Axolotl, LLaMA-Factory and Unsloth templates
- ✓Built-in MLOps: experiments, model registry, metrics
- ✓Fractional NVIDIA GPUs from 12.5% to 100%
- ✓Bring your own Docker image when you outgrow the templates
# 1. Pick the ComfyUI template
# 2. Choose a GPU fraction (e.g. 25% L40S)
# 3. Click launch — pod is ready in ~60s
curl -X POST https://cloud.podstack.ai/platform/api/v1/pods \
-H "Authorization: Bearer $PODSTACK_TOKEN" \
-d '{"template":"comfyui-l40s","gpu_fraction":0.25}'Use cases
Fine-tune an open model
Launch an Unsloth/LoRA template, point it at your dataset, ship.
Serve a model behind an API
vLLM template with autoscaling in a click.
Notebook + experiment tracking
Reproducible research env with metrics and artifacts logged automatically.
When to use QuickPods
- ✓You want to start building, not configuring infra
- ✓You need MLOps without wiring five tools together
- ✓Your workload fits on a GPU fraction and you want lower cost
How to use
- 1
Pick a template
Open QuickPods and choose a ready-made template — vLLM, Unsloth, ComfyUI, PyTorch and more — from the catalog. No CLI, no Dockerfiles.
- 2
Choose your GPU fraction
Select how much GPU you need, from 25% to 100%, powered by podvirt — you only pay for the slice you use.
- 3
Start your pod
Click start and your pod boots in seconds into a ready-to-use Jupyter notebook with the template's software pre-installed. Start building right away.
Frequently asked questions
What is QuickPods?
QuickPods is PodStack’s one-click deployment product for production-ready AI stacks. You pick a template - vLLM, PyTorch, ComfyUI, Unsloth and more - and it launches on a fractional or full NVIDIA GPU with MLOps tooling already wired in.
Which GPUs can I deploy QuickPods on?
QuickPods run on NVIDIA A10G 24GB, L40S 48GB, A100 40GB, A100 80GB and H100 80GB. Every GPU supports fractional allocation from 12.5% up to 100% through PodVirt, so you only pay for the slice you need.
How does billing work for QuickPods?
Billing is per-minute and scales linearly with the GPU fraction you allocate, and there are zero egress fees. Pricing is sales-led - contact sales@podstack.ai for a quote tailored to your workload.
What MLOps features come built in?
QuickPods ship with experiment tracking, a model registry, real-time cost tracking and metrics logging, so you can go from template to a monitored deployment without stitching tools together.
Can I bring my own Docker image instead of a template?
Yes. QuickPods start from one-click templates, but you can launch any OCI-compatible Docker image. The templates are a fast path, not a limitation.
Is PodStack built on OpenStack or Kubernetes?
No. PodStack’s control plane, scheduler and PodVirt virtualisation layer are proprietary and purpose-built for fractional GPU sharing - not OpenStack and not vanilla Kubernetes. That is what makes 12.5% GPU allocation and sub-second scaling possible.