oiyaiOpen console
ELASTIC GPU COMPUTE / DEVELOPER PREVIEW01 — DOCKER-LIKE. GPU-POWERED.

Elastic GPU.Docker-like.Simple.

OIY / ELASTIC COMPUTESCALE UP. SCALE TO ZERO.
BUILT FOR BUILDERS

Your image.
Your GPU.
Ready to run.

Start building
CONTAINERS, WITHOUT THE COMPLEXITY.

Run containers on elastic GPU compute.
Scale to zero when idle. Pay for usage.

Deploy a container
02 / THE PLATFORMFROM IDEA TO EXECUTION

A little less setup.
A lot more building.

Deploy your image, choose GPU resources, and run. Scale compute around your workload with idle scale-to-zero and usage-based billing.

01

Docker-like simplicity.

Bring an image, set your command and environment, and launch a GPU container through the console or API.

Explore compute
02

Elastic resources.

Adjust GPU count, memory, CPU, and RAM through one service workflow. Keep the endpoint and persistent data as compute changes.

Scale your resources
03

Scale to zero.

Release idle compute automatically, then wake it on demand. Persistent storage stays available, and task tracking protects active jobs.

Explore scale-to-zero
04

Pay for usage.

Compute is metered while allocated, with second-level accounting. Review resource prices, usage history, and storage charges in one place.

Understand billing
03 / THE COMPUTE COLLECTIONBUILT AROUND YOUR WORKLOAD

Small experiments.
Expansive thinking.

From partitioned GPUs to full-device profiles. Find the right amount of headroom for what you’re building.

NVIDIA / BLACKWELL
24 / 48 / 96GB
MEMORY PER DEVICE / SUPPORTED PROFILE

MIG partitions or a full GPU. Choose the memory your workload needs.

Read the GPU guide

Profiles describe supported configurations. Live inventory, placement, and pricing determine availability. Check the console ↗

04 / ONE SIMPLE CONTAINER WORKFLOWCREATE → BUILD → PAUSE → RESUME
RUN → SCALE → ZERO → RESUME

Run. Scale.
Get on with it.

The container workflow you already understand, backed by elastic GPU compute.

  1. 01

    Bring your environment

    Choose a template or use your own container image. Set resources, a command, and environment variables.

  2. 02

    Scale when you need more

    Change GPU, CPU, and memory through one service action. Your persistent workspace and endpoint stay associated with the service.

  3. 03

    Scale to zero when idle

    Let idle sleep release compute, then wake it on demand. Keep persistent storage and protect long-running work with activity tracking.

Follow the quickstart
05 / FOR THE WAY YOU BUILDCONSOLE / API / PYTHON / MCP

Your container.
A simple API.

Call the API directly. Protect a Python task. Let an MCP client help manage your workspace. It all connects to the same services.

Explore the developer guides
SDK + CLI + MCP: SOURCE DISTRIBUTION
PACKAGE REGISTRY RELEASES NOT YET AVAILABLE
service.json / CONTAINER CONFIG
{
  "name": "my-gpu-container",
  "region": "standard",
  "image": "your-registry/your-image:tag",
  "resources": {
    "gpuCount": 1,
    "gpuModel": "rtx-pro-6000-blackwell",
    "gpuMemoryGb": 96,
    "cpu": 8,
    "memoryGb": 32
  },
  "sleepAfterMinutes": 15
}
Replace the image and confirm available resources.Run the HTTP example ↗
OIY / THE ELASTIC CONTAINER LIFECYCLESCROLL TO EXPLORE

Elastic compute
with a simple workflow.

01 / DEPLOY

Start with a container.

Bring an image and a command. Pick GPU resources and launch through the console or API.

Deploy a container
YOUR IMAGE / ONE CONTAINER WORKLOAD
06 / A FEW GOOD QUESTIONSCLARITY BEFORE COMPUTE

Good to
know.

Run container-based AI workloads: model experiments, inference services, notebooks, and training jobs. You choose the image, resources, environment, and persistent workspace. Workload compatibility depends on your image and the available runtime.

07 / YOUR NEXT CHAPTEROIY AI — DEVELOPER PREVIEW

What will you
make possible?

Deploy your container Read the docs