A template is a container image with its settings, ready to deploy as an instance or a Run Job. This page explains official templates, your own templates, and what happens when you deploy from one.
Official templates are maintained environments, such as PyTorch, vLLM and SGLang, with SSH and JupyterLab included. Each lists its versions, newest first; a new deploy uses the newest unless you choose another. Each official template runs on the GPU vendor its image is built for: ROCm images on AMD GPUs, CUDA images on NVIDIA GPUs.
A template you create saves an image and every launch setting: compute, container disk, network volumes, start command, environment variables, secrets, public ports and the tmux setting. It's Private to you, or shared with your Organization so every member can deploy it, secrets included.
When you deploy from a template, anything you change on the deploy page overrides the template's value for that deploy. Environment variables and secrets merge by name. Instances and Run Jobs keep their settings when the template is later edited or deleted.