Deploying an endpoint

You can deploy a dedicated endpoint in the dashboard with a guided wizard, or via the management API at https://gateway.omnia-voice.com/v1 using your workspace API key.

From the dashboard

Open Dedicated → Deploy

In the dashboard, go to Dedicated and start a new deployment.

Choose model & hardware

Select the base model, then the region, GPU type, and GPUs per replica. Only combinations that are actually available are shown.

Set autoscaling

Choose the minimum and maximum number of replicas, and name the endpoint.

Confirm funds and deploy

The wizard shows the hourly rate. You need enough prepaid balance to cover the minimum runway; if not, it tells you the exact shortfall to add. Deploy when ready.

Discover available hardware

Before creating an endpoint, list the GPU types, regions, and flavors you can deploy against with GET /v1/dedicated/templates. Deploy only against combinations returned here.

curl https://gateway.omnia-voice.com/v1/dedicated/templates \
  -H "Authorization: Bearer $OMNIA_API_KEY"

import requests

BASE = "https://gateway.omnia-voice.com/v1"
headers = {"Authorization": f"Bearer {OMNIA_API_KEY}"}

templates = requests.get(f"{BASE}/dedicated/templates", headers=headers).json()

If a GPU count you need (above 8 per replica) or a region isn’t available in the templates, the dashboard shows “contact sales”; larger and custom deployments are arranged directly.

Create an endpoint

Create an endpoint with a POST to /v1/dedicated.

curl https://gateway.omnia-voice.com/v1/dedicated \
  -H "Authorization: Bearer $OMNIA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "prod-qwen",
    "description": "Production Qwen3-32B endpoint",
    "modelName": "Qwen/Qwen3-32B",
    "flavorName": "<flavor>",
    "gpuType": "<gpu-type>",
    "gpuCount": 1,
    "region": "<region>",
    "minReplicas": 1,
    "maxReplicas": 2
  }'

import requests

BASE = "https://gateway.omnia-voice.com/v1"
headers = {
    "Authorization": f"Bearer {OMNIA_API_KEY}",
    "Content-Type": "application/json",
}

resp = requests.post(
    f"{BASE}/dedicated",
    headers=headers,
    json={
        "name": "prod-qwen",
        "description": "Production Qwen3-32B endpoint",
        "modelName": "Qwen/Qwen3-32B",
        "flavorName": "<flavor>",
        "gpuType": "<gpu-type>",
        "gpuCount": 1,
        "region": "<region>",
        "minReplicas": 1,
        "maxReplicas": 2,
    },
)
created = resp.json()  # -> { "id": "<endpoint-id>" }

A successful create returns just the new endpoint’s id:

{ "id": "<endpoint-id>" }

The endpoint provisions asynchronously. Fetch it with GET /v1/dedicated/{id} to read its routing key and current status once it’s ready:

{
  "id": "<endpoint-id>",
  "name": "prod-qwen",
  "modelName": "Qwen/Qwen3-32B",
  "routingKey": "<routing-key>",
  "minReplicas": 1,
  "maxReplicas": 2,
  "enabled": true,
  "status": "deploying"
}

Once running, route to the endpoint by passing model = dedicated/<routing-key> on the inference API; see Managing endpoints.

Fields

Field	Required	Description
`name`	Yes	Human-readable name for the endpoint.
`modelName`	Yes	The base model to serve (a catalog id, e.g. `Qwen/Qwen3-32B`).
`description`	No	Free-text description.
`flavorName`	No	Hardware flavor, from the templates endpoint.
`gpuType`	No	GPU type to run on, from the templates endpoint.
`gpuCount`	No	GPUs per replica.
`region`	No	Region to deploy in, from the templates endpoint.
`minReplicas`	No	Lower bound for autoscaling (warm capacity).
`maxReplicas`	No	Upper bound for autoscaling (throughput/spend cap).
`customWeightsId`	No	Serve a fine-tuned model artifact by its id.
`fineTuningJobId`	No	Alternative to `customWeightsId`: serve the output of a completed fine-tuning job.

To serve a model you trained, set either customWeightsId (the model artifact id) or fineTuningJobId (a completed job) instead of relying on the base model alone. See Deploying a fine-tuned model.

Deploying starts billing for GPU time and requires prepaid balance covering the minimum runway (by default one hour). If the balance is short, the create call is refused with the exact shortfall amount. If the wallet is later exhausted while running, the endpoint is stopped automatically.

Need more than self-serve offers?

If you need more GPUs per replica than are listed, or a region that isn’t available, contact sales; larger and custom deployments are arranged directly.

Manage your endpoint

Call, scale, disable, and delete a running endpoint.

Get started

Capture

Observe

Grade & Judge

Evaluate

Improve

Run at scale

Platform

Deploying an endpoint

From the dashboard

Discover available hardware

Create an endpoint

Fields

Need more than self-serve offers?

Manage your endpoint

​From the dashboard

​Discover available hardware

​Create an endpoint

​Fields

​Need more than self-serve offers?

Manage your endpoint

From the dashboard

Discover available hardware

Create an endpoint

Fields

Need more than self-serve offers?