Get started

Concepts

The building blocks of Featherless GPU Cloud and how they fit together.

This page defines the terms you'll meet in the console, the CLI and these docs, and helps you choose between an instance and a Run Job.

Terms

Instance

A container on dedicated GPU or CPU compute. Use it for development, notebooks and services.

Run Job

A command that runs on temporary compute and ends when the command exits. You pay only while it runs. Use it for training, batch processing and evaluations.

Template

A container image with its settings, ready to deploy. Use an official template such as PyTorch or vLLM, or save your own and share it with your organization.

Network volume

Persistent storage you mount at a path you choose. It keeps its data until you delete it, and several instances and Run Jobs can mount it at once.

Container disk

Temporary storage for the container's files. It's erased when the instance stops or restarts.

Custom image

Any Linux container image you can pull, used instead of a template. Private images need registry credentials.

Public port

A port in your container served at its own HTTPS address, so you can reach a web app or API without SSH.

API key

A secret that lets the CLI or your scripts act as you. Its scopes limit what it can do.

Organization

The account that owns your instances, Run Jobs, volumes, templates and credit. Every member works with the same resources.

Instance or Run Job?

You want toUse
Work interactively in a terminal, notebook or IDEAn instance
Serve a model or web app continuouslyAn instance with a public port
Run a script once and collect its outputA Run Job
Run many independent tasksRun Jobs, one per task
Pay only while your command runsA Run Job

Next steps

Quickstart

Deploy your first instance.

Submit a Run Job

Run a command on temporary compute.