class LittleGhost::Model

Model is the configured connection between an agent role and a provider. It keeps provider choice and defaults out of the agent class that uses them.

It merges profile settings into every ModelRequest, validates attachment modalities declared in metadata, lets providers prepare capability-sensitive requests, and delegates the normalized stream to the provider.

Attributes

Provider object and name, provider model ID, default settings, normalized metadata, and logical application role.

Provider object and name, provider model ID, default settings, normalized metadata, and logical application role.

Provider object and name, provider model ID, default settings, normalized metadata, and logical application role.

provider_name [R]

Provider object and name, provider model ID, default settings, normalized metadata, and logical application role.

Provider object and name, provider model ID, default settings, normalized metadata, and logical application role.

Provider object and name, provider model ID, default settings, normalized metadata, and logical application role.

Public Class Methods

new (provider:, provider_name:, id: nil, model: nil, settings: {}, metadata: {}, role: nil)

Source

# File lib/little_ghost/model.rb, line 18
def initialize(provider:, provider_name:, id: nil, model: nil, settings: {}, metadata: {}, role: nil)
  raise ArgumentError, "provider must respond to stream" unless provider.respond_to?(:stream)
  raise ArgumentError, "provider_name is required" if provider_name.nil? || provider_name.to_s.empty?
  raise ArgumentError, "model is required" if (id || model).nil? || (id || model).to_s.empty?

  @provider = provider
  @provider_name = provider_name.to_sym
  @id = (id || model)&.to_s
  @settings = settings.to_h.transform_keys(&:to_sym).freeze
  @role = role&.to_s
  profile_metadata = metadata.to_h.reject { |key, _value| IDENTITY_METADATA_KEYS.include?(key.to_s) }
  @metadata = profile_metadata.merge(
    provider: @provider_name,
    model_id: @id,
    model_role: @role
  ).freeze
end

Wraps an object that responds to stream.

Public Instance Methods

capabilities ()

Source

# File lib/little_ghost/model.rb, line 59
def capabilities
  @capabilities ||= if provider.respond_to?(:capabilities)
    provider.capabilities(metadata:)
  else
    ModelCapabilities.legacy
  end
end

Uses advertised provider capabilities, falling back to the permissive legacy contract for providers that do not advertise them.

stream (request, &block)

Source

# File lib/little_ghost/model.rb, line 39
def stream(request, &block)
  validate_input_modalities!(request)
  configured_request = ModelRequest.new(
    messages: request.messages,
    tools: request.tools,
    settings: settings.merge(request.settings),
    output_schema: request.output_schema,
    tool_choice: request.tool_choice,
    required_capabilities: request.required_capabilities,
    cancellation_token: request.cancellation_token,
    deadline: request.deadline
  )
  if provider.respond_to?(:prepare_request)
    configured_request = provider.prepare_request(configured_request, capabilities:)
  end
  provider.stream(configured_request, &block)
end

Streams request through the configured provider.

Profile settings are defaults; settings on request take precedence.