Templates determine what is running on your GPU instance when it boots. Every instance is a virtual machine that starts from the same base image — Ubuntu 22.04 with NVIDIA drivers and Docker — and app templates start a container on top of it.

​
Available templates

​
Bare Metal

“Bare Metal” is the name of the plain template: you get the base virtual machine with nothing on top. Pick this when you install your own stack. The NVIDIA driver (nvidia-smi) and Docker are already installed, so a setup script or a Docker image of your own is usually the fastest way to bring your environment.

​
PyTorch + Jupyter

JupyterLab opens in the browser with PyTorch and CUDA 12 ready. The app runs in a Docker container on the instance, so notebooks and files you create live inside that container. To move them to the host for scp or rsync, SSH in and use docker ps to find the container, then docker cp.

​
Open WebUI + Ollama

A chat interface backed by Ollama on your GPU. No models are pre-installed. Open the app, create your Open WebUI account (see Reaching browser apps), choose Pull a model (for example llama3 or qwen2.5), and start chatting once the download completes. Model downloads count against the instance disk.

​
ComfyUI

The node-based Stable Diffusion editor. Upload checkpoints and LoRAs through the app, or scp them to the instance and docker cp them into the ComfyUI models directory inside the container.

​
Reaching browser apps

App templates get a private URL under app.quantacloud.net. When your deployment is Running, the Deployments page shows an Open Application button, and the deployment’s detail view an Open button named after the app.
  • The URL is protected by your QuantaCloud login. Only the account that owns the deployment can open it.
  • JupyterLab and ComfyUI have no separate password: after the QuantaCloud login, they open directly.
  • Open WebUI also has its own sign-in. The first time you open it, create an Open WebUI account; the first account becomes the Open WebUI admin.
  • The URL stops working when the deployment is stopped. Stopping terminates the instance and deletes its disk, including everything inside the app container, so download your work first.
App templates take a few minutes longer than Bare Metal to reach Running while the container image is pulled. The deployment stays in Connecting until the app answers its health check.

​
Choosing during deployment

Templates are selected in step 1 of the deploy flow. The template determines what is available when you connect; it does not change the price, which depends only on the offer. Not sure which to pick? PyTorch + Jupyter is the best starting point for ML work, and Bare Metal if you already have a Docker image or setup script.