Templates

A template (or stack) is a packaged piece of software you apply to a running cluster — a scheduler, an ML framework, a monitoring dashboard, or your own custom stack. Templates let you go from a bare cluster to a ready-to-use environment in one step.

Browsing templates

Open Templates in the console. Each card shows an icon, a category (Scheduler, Orchestration, ML Framework, Monitoring, …), and a short description. Search and category filters narrow the list.

Templates have a visibility:

• Global / system — published by an administrator, available to everyone.
• Private — added by you, visible only to your account.

Applying a template

1. Open a template and click Apply to a cluster (single node or a full cluster are both supported).
2. Choose the target cluster and fill in any inputs the template declares.
3. Apply. The run executes and reports progress; when it completes, the software is live on the cluster.

You can watch each run's log from the template's run view.

Adding your own template

Anyone can add a template; where it's visible depends on who adds it:

• If an admin adds it, it becomes global (available to all accounts).
• If a user adds it, it's private to that user.

Add a template two ways:

• Upload a .zip of the template, or
• Provide a download URL to a .zip.

Template package format

A template zip contains the module's *.tf files. It may optionally include a manifest.json at the root describing metadata (title, description, category, inputs). If present, the manifest populates the catalog entry; if absent, sensible defaults are derived from the module.

my-template.zip
├── manifest.json        # optional metadata (shallowest one wins)
├── main.tf              # the module
├── variables.tf
└── outputs.tf

A manifest.json looks like:

{
  "title": "JupyterLab",
  "description": "A JupyterLab server exposed on the cluster.",
  "category": "ML Framework",
  "icon": "🪐",
  "applicable_profiles": ["gpu-l40s", "gpu-hgx-h100"],
  "inputs": [
    {
      "name": "port",
      "type": "number",
      "required": false,
      "default": 8888,
      "description": "Port JupyterLab listens on."
    },
    {
      "name": "conda_env",
      "type": "string",
      "required": true,
      "description": "Name of the conda environment to launch in."
    }
  ]
}

Field	Purpose
title / description	Shown on the catalog card and detail page.
category	Filter bucket (Scheduler, Orchestration, ML Framework, Monitoring).
icon	Emoji or an image URL (see Icons).
applicable_profiles	Restrict which hardware profiles the template targets; omit for all.
inputs[]	Parameters the apply form collects: name, type, required, default, description.

How an apply runs

Applying a template runs its module against the target cluster as a tracked run. The platform injects the cluster's context (node addresses, an access identity, and platform variables such as the cluster id and public domain) so the module can wire itself to the running cluster. You watch progress in the run view; when the run reaches APPLIED, the software is live. A run that declares an exposed web port automatically gets the per-cluster public HTTPS URL described in Access & Networking.

Start from an existing template

You can download the .zip of any globally available template, modify it, and re-upload it as your own private template. This is the easiest way to build a custom stack.

Icons

Templates show an icon so they're easy to scan:

• Use an emoji, or
• Provide an image URL (a CDN link) for a logo.

Icons render in a fixed, uniform slot — images are fit inside without cropping, so every card lines up regardless of the source image's aspect ratio. Admins can change the icon of any template; users can change the icon of their own private templates.

Managing templates

Action	Who
Apply to a cluster	Anyone with access to the cluster
Download the .zip	Anyone (for globally available templates)
Change the icon	Admins (any) / owners (their private ones)
Delete	Admins (any custom) / owners (their private ones)

API surface: GET /v1/stacks, GET /v1/stacks/{name}, GET /v1/stacks/{name}/download, POST /v1/stacks, PATCH /v1/stacks/{name}, DELETE /v1/stacks/{name}.

---

Next: Images.