What is AiQu?

AiQu runs Docker containers on shared GPU and CPU machines.

You bring an image and say how much you need. AiQu finds a machine that can take it, starts your container there, gives you a terminal and a URL into it, and stops it when your runtime runs out. What you pay for is the time your container actually held those resources.

That is the entire proposition. It is deliberately not a Kubernetes, not a notebook service, and not an MLOps pipeline. It is a way to get a container onto a GPU without owning the GPU.

Who it is for

Someone with a model to train or an experiment to run. You have a Docker image, or you are happy with one of the prepared ones, and you need a few hours on a real GPU. You do not want to learn a scheduler.

A team that owns GPUs and wants other people to use them. You put your machines in as nodes, group them into queues, and hand out access per user group. The machines stay yours; the scheduling and the accounting are not your problem any more.

An organisation running several of both. Each team gets a tenant. Tenants cannot see each other’s jobs, users, storage or nodes.

What you get

• Any image. Docker Hub, or a private registry your tenant has configured.
• A shell. A terminal in the browser, or SSH from your own machine.
• A URL. Any port your container listens on can be proxied out, publicly or privately, with an IP filter if you want one.
• Storage that survives. Buckets you mount into the container. Everything outside a mount is gone when the job ends.
• A bill you can read. Tokens, charged per minute of held resources, with a transaction history per job.

What it is not

Containers here are temporary by default

A job has a runtime. When the runtime expires the container is stopped and removed. The filesystem goes with it. This surprises people once, and only once - see Files and storage.

AiQu does not restart your job when it crashes, does not autoscale, and does not keep a container alive because it is still doing something useful. If you need more time you extend the job before it expires.

Where the machines are

A node is a machine that has joined a tenant. It might be a server in a rack, a workstation under someone’s desk, or a box at a customer site. Nodes connect outward to the central server over a tunnel, so a node does not need to be reachable from the internet - it only needs to be able to reach AiQu.

This is why a node can be offline. When it is, the queues that depend on it have nothing to schedule onto, and jobs sit waiting. The System page is where you check that.