Deploy your first container
Choose an image, launch GPU compute, and configure scale-to-zero.
Before you begin
Open the Oiy AI console, create an account or sign in, and verify your email. New accounts begin with zero credit. Check Billing for a balance and an available top-up method. Creating compute needs a verified account, sufficient balance, and an enabled, priced placement with compatible capacity.
If no placement is available, you can still explore templates and the documentation. A supported GPU profile does not guarantee available hardware.
1. Choose a starting point
Open Templates and choose a built-in starter, or open Services → New service and supply a container image. Review the image, startup command, required environment variables, and exposed port.
Use an image that includes your workload's dependencies. The platform does not automatically install arbitrary Python packages into a custom image. The common custom-container runtime expects /bin/sh, OpenSSH, and coreutils. For a private repository, configure image-pull access first.
2. Configure resources and placement
Choose Standard or Low-price, then select a geography or Automatic. Choose CPU-only compute or a GPU model with a supported memory profile. Review CPU, system memory, workspace storage, and the displayed price before creating the service.
Compute profiles and placement explain the available choices.
3. Decide what should persist
Save durable files under /workspace. A service-owned volume is removed when its service is deleted. Choose independent storage when the files need to outlive the service, or attach an existing independent volume in its original placement.
Container-local files outside the persistent mount are not a durable workspace. Read the storage guide.
4. Create and wait for readiness
Submit the configuration and open the service details. Creation acknowledges asynchronous work: queued and starting do not mean your application is ready. Inspect the service state, events, and logs. If the image serves HTTP, it must listen on the configured port and an externally reachable interface such as 0.0.0.0.
5. Work, then pause
Use the service's authenticated endpoint or the browser launch control when available. For long jobs, use an activity lease so inactivity does not put a real task to sleep.
When finished, pause the service. Compute billing ends after release is confirmed; provisioned storage remains billable. To resume, wake the service and wait for it to run. The workspace remains; processes restart.