Setting up an embedding model

An embedding model entity configures a provider, its connection settings, and model options. You can create embedding models from the admin UI or via the admin API.

Entity configuration

{
  "id": "embedding-model-entity-id",
  "name": "My Embedding Model",
  "description": "An OpenAI embedding model",
  "provider": "openai",
  "config": {
    "connection": {
      "base_url": "https://api.openai.com/v1",
      "token": "sk-xxx",
      "timeout": 180000
    },
    "options": {
      "model": "text-embedding-3-small"
    }
  },
  "models": {
    "include": [],
    "exclude": []
  }
}

Connection settings

Parameter	Type	Default	Description
base_url	string	Provider-specific	API base URL
token	string	—	API token. Supports comma-separated values for round-robin rotation
timeout	number	180000	Request timeout in milliseconds

For Azure OpenAI (legacy mode), additional fields are available:

Parameter	Type	Description
resource_name	string	Azure resource name
deployment_id	string	Deployment/model ID
api_version	string	API version
api_key	string	API key (alternative to bearer token)

Model options

The options object contains provider-specific settings. The model field is common to all providers:

Parameter	Type	Description
model	string	The embedding model name (provider-specific default if not set)

Any additional fields in options are forwarded to the provider API as-is. This allows passing provider-specific parameters like dimensions for OpenAI models.

Model constraints

You can restrict which models consumers are allowed to use with regex patterns:

{
  "models": {
    "include": ["text-embedding-3-.*"],
    "exclude": [".*ada.*"]
  }
}

Model constraints are also checked against API key and user metadata (ai_models_include / ai_models_exclude). See per API key / per user restrictions for details.

Admin API

The embedding model entity is available at:

ai-gateway.extensions.cloud-apim.com/v1/embedding-models

Standard CRUD operations are supported (GET, POST, PUT, DELETE).