Gemini.Types.ModelArmorConfig (GeminiEx v0.11.0)

Copy Markdown View Source

Configuration for Model Armor integrations.

This feature is only supported in Vertex AI, not the Gemini Developer API.

Model Armor allows you to apply content filtering templates managed separately from your generation requests. This provides centralized policy management for content moderation.

Important

If model_armor_config is provided, safety_settings must NOT be provided. These two options are mutually exclusive.

Fields

  • prompt_template_name - Resource name of the Model Armor template to apply to prompt content (optional)
  • response_template_name - Resource name of the Model Armor template to apply to response content (optional)

Example

config = %Gemini.Types.ModelArmorConfig{
  prompt_template_name: "projects/my-project/locations/us-central1/templates/prompt-filter",
  response_template_name: "projects/my-project/locations/us-central1/templates/response-filter"
}

# Use in generate request (Vertex AI only)
Gemini.generate("Hello world", model_armor_config: config)

Summary

Types

t()

Model Armor configuration.

Functions

from_api(config)

Creates a ModelArmorConfig from API response.

new(opts \\ [])

Creates a new ModelArmorConfig struct.

to_api(config)

Converts ModelArmorConfig to API format (camelCase keys).

validate_exclusivity(arg1, safety_settings, arg3)

Validates that model_armor_config and safety_settings are not both provided.

Types

t()

@type t() :: %Gemini.Types.ModelArmorConfig{
  prompt_template_name: String.t() | nil,
  response_template_name: String.t() | nil
}

Model Armor configuration.

  • prompt_template_name - Model Armor template for filtering prompt content
  • response_template_name - Model Armor template for filtering response content

Functions

from_api(config)

@spec from_api(map() | nil) :: t() | nil

Creates a ModelArmorConfig from API response.

Parameters

  • data - Map from API response with camelCase string keys

new(opts \\ [])

@spec new(keyword()) :: t()

Creates a new ModelArmorConfig struct.

Parameters

  • opts - Keyword list with configuration options:
    • :prompt_template_name - Template name for prompt filtering
    • :response_template_name - Template name for response filtering

Examples

config = Gemini.Types.ModelArmorConfig.new(
  prompt_template_name: "projects/my-project/locations/us-central1/templates/t1"
)

to_api(config)

@spec to_api(t() | nil) :: map() | nil

Converts ModelArmorConfig to API format (camelCase keys).

Examples

config = %ModelArmorConfig{
  prompt_template_name: "t1",
  response_template_name: "t2"
}

ModelArmorConfig.to_api(config)
#=> %{"promptTemplateName" => "t1", "responseTemplateName" => "t2"}

validate_exclusivity(arg1, safety_settings, arg3)

@spec validate_exclusivity(t() | nil, list() | nil, :gemini | :vertex_ai) ::
  :ok | {:error, String.t()}

Validates that model_armor_config and safety_settings are not both provided.

This is a helper for request building - these options are mutually exclusive.

Parameters

  • model_armor_config - The model armor config (or nil)
  • safety_settings - The safety settings list (or nil/empty)
  • api_type - Current API type (:gemini or :vertex_ai)

Returns

  • :ok - Validation passed
  • {:error, reason} - Validation failed

Examples

validate_exclusivity(%ModelArmorConfig{...}, [], :vertex_ai)
#=> :ok

validate_exclusivity(%ModelArmorConfig{...}, [%SafetySetting{}], :vertex_ai)
#=> {:error, "model_armor_config and safety_settings are mutually exclusive"}

validate_exclusivity(%ModelArmorConfig{...}, nil, :gemini)
#=> {:error, "model_armor_config is only supported in Vertex AI"}