Radius.AI/models@2025-08-01-preview

Description

The Radius.AI/models Resource Type deploys an LLM inference model endpoint. To deploy a model, add a models resource to the application definition Bicep file. The platform-engineer recipe binds this developer-facing contract to a managed cloud model service without exposing any module details to the app.

resource model 'Radius.AI/models@2025-08-01-preview' = {
  name: 'model'
  properties: {
    environment: environment
    application: myApplication.id
    model: 'gpt-5-mini'
  }
}

To connect your container to the model, create a connection from the Container resource to the model as shown below.

resource frontend 'Radius.Compute/containers@2025-08-01-preview' = {
  name: 'frontend'
  properties: {
    application: myApplication.id
    environment: environment
    containers: {
      frontend: {
        image: 'frontend:1.25'
      }
    }
    connections: {
      llm: {
        source: model.id
      }
    }
  }
}

On compatible Kubernetes Container Recipes, the connection injects environment variables into the container for the model properties. The variables are named CONNECTION_<CONNECTION-NAME>_<PROPERTY-NAME>. In this example the connection name is llm so the environment variables will be:

• CONNECTION_LLM_MODEL
• CONNECTION_LLM_ENDPOINT
• CONNECTION_LLM_APIKEY (secret-backed)

With Radius control-plane support from radius-project/radius#12709 and Kubernetes Container Recipe support from resource-types-contrib#300 or later, the same connection injects apiKey through a Kubernetes secret reference. For custom, older, or mixed-version Kubernetes deployments, use model.properties.secrets.name as the secretName and apiKey as the key in an explicitly authored secretKeyRef.

Top-Level Properties

Object Properties

connections

secrets