raif 1.3.0 → 1.4.0

data/lib/generators/raif/install/templates/initializer.rb

@@ -10,6 +10,22 @@ Raif.configure do |config|
   # Whether OpenAI embedding models are enabled.
   # config.open_ai_embedding_models_enabled = ENV["OPENAI_API_KEY"].present?
 
+  # The base URL for OpenAI API requests.
+  # Set this if you want to use the OpenAI adapter with a different provider (e.g. for using Azure instead of OpenAI)
+  # config.open_ai_base_url = "https://api.openai.com/v1"
+
+  # The base URL for OpenAI embedding API requests.
+  # Set this if you want to use a different provider for embeddings (e.g. Ollama, vLLM, or other OpenAI-compatible APIs)
+  # config.open_ai_embedding_base_url = "https://api.openai.com/v1"
+
+  # When set, this will be included as an api-version parameter in any OpenAI API requests (e.g. for using Azure instead of OpenAI)
+  # config.open_ai_api_version = nil
+
+  # The authentication header style for OpenAI API requests. Defaults to :bearer
+  # Use :bearer for standard OpenAI API (Authorization: Bearer <token>)
+  # Use :api_key for Azure OpenAI API (api-key: <token>)
+  # config.open_ai_auth_header_style = :bearer
+
   # Your Anthropic API key. Defaults to ENV["ANTHROPIC_API_KEY"]
   # config.anthropic_api_key = ENV["ANTHROPIC_API_KEY"]
 
@@ -40,24 +56,56 @@ Raif.configure do |config|
   # The site URL to include in OpenRouter API requests headers. Optional.
   # config.open_router_site_url = "https://myapp.com"
 
+  # Your Google AI API key. Defaults to ENV["GOOGLE_AI_API_KEY"].presence || ENV["GOOGLE_API_KEY"]
+  # config.google_api_key = ENV["GOOGLE_AI_API_KEY"].presence || ENV["GOOGLE_API_KEY"]
+
+  # Whether Google models are enabled.
+  # config.google_models_enabled = ENV["GOOGLE_API_KEY"].present?
+
   # The default LLM model to use. Defaults to "open_ai_gpt_4o"
   # Available keys:
-  #   open_ai_gpt_4_1
-  #   open_ai_gpt_4_1_mini
-  #   open_ai_gpt_4_1_nano
   #   open_ai_gpt_4o_mini
   #   open_ai_gpt_4o
   #   open_ai_gpt_3_5_turbo
+  #   open_ai_gpt_4_1
+  #   open_ai_gpt_4_1_mini
+  #   open_ai_gpt_4_1_nano
+  #   open_ai_o1
+  #   open_ai_o1_mini
+  #   open_ai_o3
+  #   open_ai_o3_mini
+  #   open_ai_o4_mini
+  #   open_ai_gpt_5
+  #   open_ai_gpt_5_mini
+  #   open_ai_gpt_5_nano
+  #   open_ai_responses_gpt_4o_mini
+  #   open_ai_responses_gpt_4o
+  #   open_ai_responses_gpt_3_5_turbo
   #   open_ai_responses_gpt_4_1
   #   open_ai_responses_gpt_4_1_mini
   #   open_ai_responses_gpt_4_1_nano
-  #   open_ai_responses_gpt_4o_mini
-  #   open_ai_responses_gpt_4o
-  #   open_ai_gpt_3_5_turbo
+  #   open_ai_responses_o1
+  #   open_ai_responses_o1_mini
+  #   open_ai_responses_o3
+  #   open_ai_responses_o3_mini
+  #   open_ai_responses_o4_mini
+  #   open_ai_responses_gpt_5
+  #   open_ai_responses_gpt_5_mini
+  #   open_ai_responses_gpt_5_nano
+  #   open_ai_responses_o1_pro
+  #   open_ai_responses_o3_pro
+  #   anthropic_claude_4_sonnet
+  #   anthropic_claude_4_5_sonnet
+  #   anthropic_claude_4_opus
+  #   anthropic_claude_4_1_opus
   #   anthropic_claude_3_7_sonnet
   #   anthropic_claude_3_5_sonnet
   #   anthropic_claude_3_5_haiku
   #   anthropic_claude_3_opus
+  #   bedrock_claude_4_sonnet
+  #   bedrock_claude_4_5_sonnet
+  #   bedrock_claude_4_opus
+  #   bedrock_claude_4_1_opus
   #   bedrock_claude_3_5_sonnet
   #   bedrock_claude_3_7_sonnet
   #   bedrock_claude_3_5_haiku
@@ -66,10 +114,21 @@ Raif.configure do |config|
   #   bedrock_amazon_nova_lite
   #   bedrock_amazon_nova_pro
   #   open_router_claude_3_7_sonnet
-  #   open_router_llama_3_3_70b_instruct
-  #   open_router_llama_3_1_8b_instruct
-  #   open_router_gemini_2_0_flash
   #   open_router_deepseek_chat_v3
+  #   open_router_deepseek_v3_1
+  #   open_router_gemini_2_0_flash
+  #   open_router_gemini_2_5_pro
+  #   open_router_grok_4
+  #   open_router_llama_3_1_8b_instruct
+  #   open_router_llama_3_3_70b_instruct
+  #   open_router_llama_4_maverick
+  #   open_router_llama_4_scout
+  #   open_router_open_ai_gpt_oss_120b
+  #   open_router_open_ai_gpt_oss_20b
+  #   google_gemini_2_5_pro
+  #   google_gemini_2_5_flash
+  #   google_gemini_3_0_pro
+  #   google_gemini_3_0_flash
   #
   # config.default_llm_model_key = "open_ai_gpt_4o"
 
@@ -118,10 +177,14 @@ Raif.configure do |config|
   # If you want to use a custom controller that inherits from Raif::ConversationEntriesController, you can set it here.
   # config.conversation_entries_controller = "Raif::ConversationEntriesController"
 
+  # The default maximum number of conversation entries to include in LLM messages. Defaults to 50.
+  # Set to nil to include all entries. Each conversation can override this with its own llm_messages_max_length attribute.
+  # config.conversation_llm_messages_max_length_default = 50
+
   # The method to call to get the current user. Defaults to :current_user
   # config.current_user_method = :current_user
 
-  # The agent types that are available. Defaults to Set.new(["Raif::Agents::ReActAgent", "Raif::Agents::NativeToolCallingAgent"])
+  # The agent types that are available. Defaults to Set.new(["Raif::Agents::NativeToolCallingAgent"])
   # If you want to use custom agent types that inherits from Raif::Agent, you can add them here.
   # config.agent_types += ["MyAgent"]
 
@@ -138,6 +201,11 @@ Raif.configure do |config|
   # Use this to globally disable requests to LLM APIs.
   # config.llm_api_requests_enabled = true
 
+  # Timeout settings for LLM API requests (in seconds). All default to nil (use Faraday defaults).
+  # config.request_open_timeout = nil  # Time to wait for a connection to be opened
+  # config.request_read_timeout = nil  # Time to wait for data to be read
+  # config.request_write_timeout = nil # Time to wait for data to be written
+
   # The default LLM model to use for LLM-as-judge evaluations.
   # If not set, falls back to the default_llm_model_key.
   # config.evals_default_llm_judge_model_key = ENV["RAIF_EVALS_DEFAULT_LLM_JUDGE_MODEL_KEY"].presence

data/lib/generators/raif/task/templates/task.rb.tt

@@ -17,7 +17,7 @@
     # Define any attributes that are needed for the task.
     # You can then pass them when running the task and they will be available in build_prompt:
     # Raif::Tasks::<%= class_name %>.run(your_attribute: "some value")
-    # task_run_arg :your_attribute
+    # run_with :your_attribute
     <%- if options[:response_format] == "json" -%>
 
     # Define a JSON schema that the model's response should adhere to

data/lib/raif/configuration.rb

@@ -12,6 +12,7 @@ module Raif
       :aws_bedrock_region,
       :bedrock_embedding_models_enabled,
       :conversation_entries_controller,
+      :conversation_llm_messages_max_length_default,
       :conversation_system_prompt_intro,
       :conversation_types,
       :conversations_controller,
@@ -20,17 +21,26 @@ module Raif
       :default_llm_model_key,
       :evals_default_llm_judge_model_key,
       :evals_verbose_output,
+      :google_api_key,
+      :google_models_enabled,
       :llm_api_requests_enabled,
       :llm_request_max_retries,
       :llm_request_retriable_exceptions,
       :model_superclass,
       :open_ai_api_key,
+      :open_ai_api_version,
+      :open_ai_auth_header_style,
+      :open_ai_base_url,
+      :open_ai_embedding_base_url,
       :open_ai_embedding_models_enabled,
       :open_ai_models_enabled,
       :open_router_api_key,
       :open_router_models_enabled,
       :open_router_app_name,
       :open_router_site_url,
+      :request_open_timeout,
+      :request_read_timeout,
+      :request_write_timeout,
       :streaming_update_chunk_size_threshold,
       :task_creator_optional,
       :task_system_prompt_intro,
@@ -43,7 +53,7 @@ module Raif
     alias_method :aws_bedrock_titan_embedding_models_enabled=, :bedrock_embedding_models_enabled=
 
     def initialize
-      @agent_types = Set.new(["Raif::Agents::ReActAgent", "Raif::Agents::NativeToolCallingAgent"])
+      @agent_types = Set.new(["Raif::Agents::NativeToolCallingAgent"])
       @anthropic_api_key = default_disable_llm_api_requests? ? "placeholder-anthropic-api-key" : ENV["ANTHROPIC_API_KEY"]
       @bedrock_models_enabled = false
       @anthropic_models_enabled = ENV["ANTHROPIC_API_KEY"].present?
@@ -54,6 +64,7 @@ module Raif
       @bedrock_embedding_models_enabled = false
       @task_system_prompt_intro = "You are a helpful assistant."
       @conversation_entries_controller = "Raif::ConversationEntriesController"
+      @conversation_llm_messages_max_length_default = 50
       @conversation_system_prompt_intro = "You are a helpful assistant who is collaborating with a teammate."
       @conversation_types = Set.new(["Raif::Conversation"])
       @conversations_controller = "Raif::ConversationsController"
@@ -62,15 +73,24 @@ module Raif
       @default_llm_model_key = default_disable_llm_api_requests? ? :raif_test_llm : (ENV["RAIF_DEFAULT_LLM_MODEL_KEY"].presence || "open_ai_gpt_4o")
       @evals_default_llm_judge_model_key = ENV["RAIF_EVALS_DEFAULT_LLM_JUDGE_MODEL_KEY"].presence
       @evals_verbose_output = false
+      google_api_key = ENV["GOOGLE_AI_API_KEY"].presence || ENV["GOOGLE_API_KEY"]
+      @google_api_key = default_disable_llm_api_requests? ? "placeholder-google-api-key" : google_api_key
+      @google_models_enabled = @google_api_key.present?
       @llm_api_requests_enabled = !default_disable_llm_api_requests?
       @llm_request_max_retries = 2
       @llm_request_retriable_exceptions = [
         Faraday::ConnectionFailed,
         Faraday::TimeoutError,
         Faraday::ServerError,
+        Net::ReadTimeout,
+        Net::OpenTimeout,
       ]
       @model_superclass = "ApplicationRecord"
       @open_ai_api_key = default_disable_llm_api_requests? ? "placeholder-open-ai-api-key" : ENV["OPENAI_API_KEY"]
+      @open_ai_api_version = nil
+      @open_ai_auth_header_style = :bearer
+      @open_ai_base_url = "https://api.openai.com/v1"
+      @open_ai_embedding_base_url = "https://api.openai.com/v1"
       @open_ai_embedding_models_enabled = ENV["OPENAI_API_KEY"].present?
       @open_ai_models_enabled = ENV["OPENAI_API_KEY"].present?
       open_router_api_key = ENV["OPEN_ROUTER_API_KEY"].presence || ENV["OPENROUTER_API_KEY"]
@@ -78,6 +98,9 @@ module Raif
       @open_router_models_enabled = @open_router_api_key.present?
       @open_router_app_name = nil
       @open_router_site_url = nil
+      @request_open_timeout = nil
+      @request_read_timeout = nil
+      @request_write_timeout = nil
       @streaming_update_chunk_size_threshold = 25
       @task_creator_optional = true
       @user_tool_types = []
@@ -103,7 +126,9 @@ module Raif
           "Raif.config.default_llm_model_key was set to #{default_llm_model_key}, but must be one of: #{Raif.available_llm_keys.join(", ")}"
       end
 
-      if Raif.embedding_model_registry.present? && !Raif.available_embedding_model_keys.include?(default_embedding_model_key.to_sym)
+      if default_embedding_model_key.present? &&
+          Raif.embedding_model_registry.present? &&
+          !Raif.available_embedding_model_keys.include?(default_embedding_model_key.to_sym)
         raise Raif::Errors::InvalidConfigError,
           "Raif.config.default_embedding_model_key was set to #{default_embedding_model_key}, but must be one of: #{Raif.available_embedding_model_keys.join(", ")}" # rubocop:disable Layout/LineLength
       end
@@ -127,6 +152,11 @@ module Raif
           "Raif.config.open_ai_api_key is required when Raif.config.open_ai_models_enabled is true. Set it via Raif.config.open_ai_api_key or ENV[\"OPENAI_API_KEY\"]" # rubocop:disable Layout/LineLength
       end
 
+      if open_ai_models_enabled && ![:bearer, :api_key].include?(open_ai_auth_header_style)
+        raise Raif::Errors::InvalidConfigError,
+          "Raif.config.open_ai_auth_header_style must be either :bearer or :api_key"
+      end
+
       if open_ai_embedding_models_enabled && open_ai_api_key.blank?
         raise Raif::Errors::InvalidConfigError,
           "Raif.config.open_ai_api_key is required when Raif.config.open_ai_embedding_models_enabled is true. Set it via Raif.config.open_ai_api_key or ENV[\"OPENAI_API_KEY\"]" # rubocop:disable Layout/LineLength
@@ -141,6 +171,11 @@ module Raif
         raise Raif::Errors::InvalidConfigError,
           "Raif.config.open_router_api_key is required when Raif.config.open_router_models_enabled is true. Set it via Raif.config.open_router_api_key or ENV['OPEN_ROUTER_API_KEY']" # rubocop:disable Layout/LineLength
       end
+
+      if google_models_enabled && google_api_key.blank?
+        raise Raif::Errors::InvalidConfigError,
+          "Raif.config.google_api_key is required when Raif.config.google_models_enabled is true. Set it via Raif.config.google_api_key or ENV['GOOGLE_API_KEY']" # rubocop:disable Layout/LineLength
+      end
     end
 
   private

data/lib/raif/engine.rb

@@ -72,6 +72,14 @@ module Raif
       end
     end
 
+    config.after_initialize do
+      next unless Raif.config.google_models_enabled
+
+      Raif.default_llms[Raif::Llms::Google].each do |llm_config|
+        Raif.register_llm(Raif::Llms::Google, **llm_config)
+      end
+    end
+
     config.after_initialize do
       next unless Raif.config.bedrock_embedding_models_enabled
 

data/lib/raif/errors/instance_dependent_schema_error.rb

@@ -0,0 +1,8 @@
+# frozen_string_literal: true
+
+module Raif
+  module Errors
+    class InstanceDependentSchemaError < StandardError
+    end
+  end
+end

data/lib/raif/errors/streaming_error.rb

@@ -3,16 +3,19 @@
 module Raif
   module Errors
     class StreamingError < StandardError
-      attr_reader :message, :type, :code, :event
+      attr_reader :type, :code, :event
 
       def initialize(message:, type:, event:, code: nil)
-        super
+        super(message)
 
-        @message = message
         @type = type
         @code = code
         @event = event
       end
+
+      def to_s
+        "[#{type}] #{super} (code=#{code}, event=#{event})"
+      end
     end
   end
 end

data/lib/raif/errors.rb

@@ -9,3 +9,4 @@ require "raif/errors/invalid_model_image_input_error"
 require "raif/errors/invalid_model_file_input_error"
 require "raif/errors/unsupported_feature_error"
 require "raif/errors/streaming_error"
+require "raif/errors/instance_dependent_schema_error"

data/lib/raif/evals/llm_judge.rb

@@ -9,8 +9,8 @@ module Raif
       # Default to JSON response format for structured output
       llm_response_format :json
 
-      task_run_arg :content_to_judge # the content to judge
-      task_run_arg :additional_context # additional context to be provided to the judge
+      run_with :content_to_judge # the content to judge
+      run_with :additional_context # additional context to be provided to the judge
 
       def default_llm_model_key
         Raif.config.evals_default_llm_judge_model_key || super

data/lib/raif/evals/llm_judges/binary.rb

@@ -4,9 +4,9 @@ module Raif
   module Evals
     module LlmJudges
       class Binary < Raif::Evals::LlmJudge
-        task_run_arg :criteria
-        task_run_arg :examples
-        task_run_arg :strict_mode
+        run_with :criteria
+        run_with :examples
+        run_with :strict_mode
 
         json_response_schema do
           boolean :passes, description: "Whether the content passes the criteria"

data/lib/raif/evals/llm_judges/comparative.rb

@@ -4,9 +4,9 @@ module Raif
   module Evals
     module LlmJudges
       class Comparative < Raif::Evals::LlmJudge
-        task_run_arg :over_content # the content to compare against
-        task_run_arg :comparison_criteria # the criteria to use when comparing content_to_judge to over_content
-        task_run_arg :allow_ties # whether to allow ties in the comparison
+        run_with :over_content # the content to compare against
+        run_with :comparison_criteria # the criteria to use when comparing content_to_judge to over_content
+        run_with :allow_ties # whether to allow ties in the comparison
 
         attr_accessor :content_a, :content_b, :expected_winner
 

data/lib/raif/evals/llm_judges/scored.rb

@@ -4,7 +4,7 @@ module Raif
   module Evals
     module LlmJudges
       class Scored < Raif::Evals::LlmJudge
-        task_run_arg :scoring_rubric # the scoring rubric to use when evaluating the content
+        run_with :scoring_rubric # the scoring rubric to use when evaluating the content
 
         json_response_schema do
           number :score, description: "Numerical score based on the rubric"

data/lib/raif/evals/llm_judges/summarization.rb

@@ -4,8 +4,8 @@ module Raif
   module Evals
     module LlmJudges
       class Summarization < Raif::Evals::LlmJudge
-        task_run_arg :original_content # the original content to evaluate the summary against
-        task_run_arg :summary # the summary to evaluate against the original content
+        run_with :original_content # the original content to evaluate the summary against
+        run_with :summary # the summary to evaluate against the original content
 
         json_response_schema do
           object :coverage do

data/lib/raif/evals/run.rb

@@ -33,6 +33,7 @@ module Raif
         output.puts "\nStarting Raif Eval Run"
         output.puts ""
         output.puts "Raif.config.default_llm_model_key: #{Raif.config.default_llm_model_key}"
+        output.puts "Raif.config.evals_default_llm_judge_model_key: #{Raif.config.evals_default_llm_judge_model_key}"
         output.puts ""
         output.puts "=" * 50
 

data/lib/raif/json_schema_builder.rb

@@ -10,6 +10,20 @@ module Raif
       @items_schema = nil
     end
 
+    # Build schema with instance context for instance-dependent schemas
+    # The block receives the instance as a parameter and has access to the builder methods
+    #
+    # @param instance [Object] The instance to use as context
+    # @param block [Proc] The block to evaluate with instance context
+    # @return [JsonSchemaBuilder] self for chaining
+    def build_with_instance(instance, &block)
+      # Evaluate the block in the context of the builder, passing the instance as parameter
+      # This allows the block to use both builder methods (string, integer, etc.)
+      # and access the instance parameter for conditional logic
+      instance_exec(instance, &block)
+      self
+    end
+
     def string(name, options = {})
       add_property(name, "string", options)
     end