Providers - DeployGPU

These endpoints are public and do not require authentication.

List Providers

GET /api/v1/providers

Returns the GPU providers DeployGPU currently supports.

Response

{
  "data": [
    { "name": "hyperstack", "type": "vm", "status": "active" },
    { "name": "masscompute", "type": "vm", "status": "active" },
    { "name": "quantacloud", "type": "vm", "status": "active" },
    { "name": "vastai", "type": "container", "status": "active" },
    {
      "name": "verda",
      "type": "vm",
      "status": "maintenance",
      "maintenance": {
        "state": "active",
        "starts_at": "2026-10-05T08:50:00Z",
        "ends_at": "2026-10-05T10:30:00Z",
        "message": "Verda is under maintenance until 10:30 UTC on Oct 5. New deployments, power actions and volume changes on Verda are paused. Running instances are not affected. A powered-off VM cannot be started until maintenance ends."
      }
    }
  ]
}

Type	Description
`vm`	Virtual machine providers. Deploy with an OS image. Supported via the API and the dashboard
`container`	Container providers (Vast.ai). Deploy with Docker images, from the dashboard only

Field	Description
`status`	`active`, or `maintenance` while new deployments on the provider are refused
`maintenance`	Present only when a maintenance window is in force or starts within 48 hours
`maintenance.state`	`scheduled` (announced, nothing is paused yet) or `active` (in force)
`maintenance.starts_at`	When the window starts or started
`maintenance.ends_at`	The expected end. `null` when no end was given or the window has run past it
`maintenance.message`	A sentence you can show to your own users

Use the name value here, not an offer’s provider.slug, as the provider field on POST /api/v1/deployments. They match for every provider except Massed Compute, whose offers report the slug massed-compute.

Provider maintenance

Providers schedule maintenance on their own platforms, and sometimes their API misbehaves. While a provider is in maintenance on DeployGPU:

Creating a deployment, shutting down or starting a VM, and creating a volume return 503 provider_maintenance with a Retry-After header. The refused request creates nothing and costs nothing.
Running deployments keep running and are billed as usual. You can still stop them. Billing ends immediately, and the instance is removed as soon as the provider accepts the request.
A VM that is powered off stays off until the maintenance ends, because start is paused too. It is billed as usual while it is off. If you need it running during a window, start it before the window begins. The maintenance object is there from 48 hours ahead.
Deleting a volume is accepted. The volume moves to deleting, is no longer billed, and is removed once the provider confirms, which can take until the maintenance ends.

status changes to maintenance when the window starts. From 48 hours before that, the maintenance object is already present with state: scheduled, so you can warn your users ahead of time. A window that runs past its expected end stays in force until the provider is working normally again. ends_at is null from then on. Check status before you deploy, or handle the 503 and retry after the number of seconds in Retry-After. That is the time until the expected end, or 600 when no end is known.

List OS Images

GET /api/v1/providers/{provider}/images

Returns available OS images for a VM provider. Use the id field as os_image_id when creating a deployment. Results are cached for 1 hour.

Query Parameters

Param	Type	Description
`instance_type`	string	Optional. Filter images available for a specific instance type

Response

{
  "data": [
    {
      "id": "ubuntu-22.04-cuda-12.4-docker",
      "displayName": "Ubuntu 22.04 + CUDA 12.4 + Docker",
      "slug": "ubuntu-22.04-cuda-12.4-docker",
      "hasCuda": true,
      "cudaVersion": "12.4",
      "hasDocker": true,
      "isDefault": false
    }
  ]
}

Field	Description
`id`	Image identifier. Pass this as `os_image_id`
`displayName`	Human-readable name
`hasCuda`	Whether CUDA drivers are pre-installed
`cudaVersion`	CUDA version if installed (for example, `"12.4"`)
`hasDocker`	Whether Docker is pre-installed
`isDefault`	Whether this is the provider’s default image

List Volume Types

GET /api/v1/providers/{provider}/volume-types

Returns available volume types for providers that support persistent storage. Use these values as volume_type when creating a volume. Results are cached for 1 hour.

Response

{
  "data": ["Cloud-SSD"]
}

Not all providers support volumes. If a provider does not, this returns an empty list.

​List Providers

​Response

​Provider maintenance

​List OS Images

​Query Parameters

​Response

​List Volume Types

​Response

List Providers

Response

Provider maintenance

List OS Images

Query Parameters

Response

List Volume Types

Response