oiyaiOpen console
OIY AI / PRODUCTCONTAINER-LEVEL ELASTIC GPU COMPUTE
DEVELOPER PREVIEW

Your container.
Elastic GPU compute.

A Docker-like way to run GPU workloads. Bring your environment, choose resources, and build. Resize when work changes. Let idle compute sleep.

Deploy a container Explore pricing
IMAGE → CONTAINER

Start with what you know.

An image, a command, and your environment. One container service to build around.

Explore the product
01 / A FAMILIAR STARTING POINTIMAGE · COMMAND · ENVIRONMENT

Bring the container.
Keep your workflow.

Start from the built-in PyTorch/JupyterLab template or use a compatible container image. Configure your command, environment and HTTP port in the console or through the API.

Deploy your first container
THE CONTAINER CONTRACT
IMAGEYour runtime and dependencies
COMMANDThe process you want to run
ENVIRONMENTApplication configuration
WORKSPACE/workspace — persistent files

Custom images must meet the runtime requirements ↗. Configure private-image access with account-owned registry credentials.

BUILT FORModel experimentsInference servicesNotebooksTraining jobs
02 / COMPUTE THAT FITS THE WORKGPU · CPU · SYSTEM MEMORY
MODEL GRAPH / ALLOCATED COMPUTE

Start with a slice.
Make room for more.

Choose a PRO 6000 MIG profile or a full GPU, alongside CPU and system RAM. Adjust your resource configuration as the workload changes.

PRO 6000 Blackwell24 / 48 GB MIG · 96 GB full
H200 · B200 · B300141 / 180 / 288 GB profiles

These are supported profile definitions. Live prices, capacity and placement determine availability. Resizing is asynchronous and can restart processes; it does not add automatic replicas.

See current resource rates
03 / IDLE COMPUTE CAN SLEEPYOUR WORKSPACE STAYS WITH YOU

Pause the compute.
Keep the progress.

Idle sleep releases compute after the configured timeout. An authenticated request can wake the service again. Keep long-running work protected with activity tracking.

Save durable files under /workspace.
Choose independent storage when files must outlive a service.
Resume with your files; processes and in-memory state start again.

Retained storage stays billable during sleep. Compute billing ends after confirmed release. Cold-start time depends on the image, initialization and available capacity.

Understand sleep and wake
COMPUTE RELEASED / WORKSPACE RETAINED
04 / A SERVICE YOU CAN WORK WITHACCESS · OBSERVE · CONTROL
ACCOUNT ACCESS / YOUR APPLICATION

Open your app.
See what it’s doing.

Reach applications through account-authenticated HTTP endpoints or browser launch when enabled. Inspect service events, startup logs and available resource metrics.

Account-scoped accessKeep endpoint access tied to your account. Application logins, such as Jupyter’s, remain separate.
One service interfaceCreate, configure, resize, restart, pause and resume through the same control API.
Connect to your application
05 / CHOOSE YOUR WAY INSAME PRODUCT, FAMILIAR TOOLS

From first click
to your own tools.

Use the console for a first deployment. Integrate the HTTP API, work from Python or the CLI, or connect an MCP client. Follow the preview setup guides for current installation and access.

01ConsoleConfigure resources and manage services.02HTTP APIBuild your own container workflow.03Python & CLIManage resources and protect active tasks.04MCPBring service tools into a compatible AI client.
DEVELOPER PREVIEW

Features described here follow the current implementation. Deployed versions, account eligibility, configured prices and regional capacity determine what you can use now.

Preview boundaries

Give your next idea
some compute.