raif 1.1.0 → 1.2.1.pre

data/lib/raif/configuration.rb

@@ -4,13 +4,13 @@ module Raif
   class Configuration
     attr_accessor :agent_types,
       :anthropic_api_key,
-      :anthropic_bedrock_models_enabled,
+      :bedrock_models_enabled,
       :anthropic_models_enabled,
       :authorize_admin_controller_action,
       :authorize_controller_action,
       :aws_bedrock_model_name_prefix,
       :aws_bedrock_region,
-      :aws_bedrock_titan_embedding_models_enabled,
+      :bedrock_embedding_models_enabled,
       :conversation_entries_controller,
       :conversation_system_prompt_intro,
       :conversation_types,
@@ -29,20 +29,27 @@ module Raif
       :open_router_models_enabled,
       :open_router_app_name,
       :open_router_site_url,
+      :streaming_update_chunk_size_threshold,
       :task_system_prompt_intro,
       :user_tool_types
 
+    alias_method :anthropic_bedrock_models_enabled, :bedrock_models_enabled
+    alias_method :anthropic_bedrock_models_enabled=, :bedrock_models_enabled=
+
+    alias_method :aws_bedrock_titan_embedding_models_enabled, :bedrock_embedding_models_enabled
+    alias_method :aws_bedrock_titan_embedding_models_enabled=, :bedrock_embedding_models_enabled=
+
     def initialize
       # Set default config
       @agent_types = Set.new(["Raif::Agents::ReActAgent", "Raif::Agents::NativeToolCallingAgent"])
       @anthropic_api_key = ENV["ANTHROPIC_API_KEY"]
-      @anthropic_bedrock_models_enabled = false
+      @bedrock_models_enabled = false
       @anthropic_models_enabled = ENV["ANTHROPIC_API_KEY"].present?
       @authorize_admin_controller_action = ->{ false }
       @authorize_controller_action = ->{ false }
       @aws_bedrock_region = "us-east-1"
       @aws_bedrock_model_name_prefix = "us"
-      @aws_bedrock_titan_embedding_models_enabled = false
+      @bedrock_embedding_models_enabled = false
       @task_system_prompt_intro = "You are a helpful assistant."
       @conversation_entries_controller = "Raif::ConversationEntriesController"
       @conversation_system_prompt_intro = "You are a helpful assistant who is collaborating with a teammate."
@@ -66,16 +73,31 @@ module Raif
       @open_router_models_enabled = ENV["OPENROUTER_API_KEY"].present?
       @open_router_app_name = nil
       @open_router_site_url = nil
+      @streaming_update_chunk_size_threshold = 25
       @user_tool_types = []
     end
 
     def validate!
+      if Raif.llm_registry.blank?
+        puts <<~EOS
+
+          !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
+          No LLMs are enabled in Raif. Make sure you have an API key configured for at least one LLM provider. You can do this by setting an API key in your environment variables or in config/initializers/raif.rb (e.g. ENV["OPENAI_API_KEY"], ENV["ANTHROPIC_API_KEY"], ENV["OPENROUTER_API_KEY"]).
+
+          See the README for more information: https://github.com/CultivateLabs/raif#setup
+          !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
+
+        EOS
+
+        return
+      end
+
       unless Raif.available_llm_keys.include?(default_llm_model_key.to_sym)
         raise Raif::Errors::InvalidConfigError,
           "Raif.config.default_llm_model_key was set to #{default_llm_model_key}, but must be one of: #{Raif.available_llm_keys.join(", ")}"
       end
 
-      unless Raif.available_embedding_model_keys.include?(default_embedding_model_key.to_sym)
+      if 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

data/lib/raif/embedding_model_registry.rb

@@ -70,7 +70,7 @@ module Raif
           default_output_vector_size: 1536,
         },
       ],
-      Raif::EmbeddingModels::BedrockTitan => [
+      Raif::EmbeddingModels::Bedrock => [
         {
           key: :bedrock_titan_embed_text_v2,
           api_name: "amazon.titan-embed-text-v2:0",

data/lib/raif/engine.rb

@@ -29,8 +29,12 @@ module Raif
     config.after_initialize do
       next unless Raif.config.open_ai_models_enabled
 
-      Raif.default_llms[Raif::Llms::OpenAi].each do |llm_config|
-        Raif.register_llm(Raif::Llms::OpenAi, **llm_config)
+      Raif.default_llms[Raif::Llms::OpenAiCompletions].each do |llm_config|
+        Raif.register_llm(Raif::Llms::OpenAiCompletions, **llm_config)
+      end
+
+      Raif.default_llms[Raif::Llms::OpenAiResponses].each do |llm_config|
+        Raif.register_llm(Raif::Llms::OpenAiResponses, **llm_config)
       end
     end
 
@@ -51,12 +55,12 @@ module Raif
     end
 
     config.after_initialize do
-      next unless Raif.config.anthropic_bedrock_models_enabled
+      next unless Raif.config.bedrock_models_enabled
 
       require "aws-sdk-bedrockruntime"
 
-      Raif.default_llms[Raif::Llms::BedrockClaude].each do |llm_config|
-        Raif.register_llm(Raif::Llms::BedrockClaude, **llm_config)
+      Raif.default_llms[Raif::Llms::Bedrock].each do |llm_config|
+        Raif.register_llm(Raif::Llms::Bedrock, **llm_config)
       end
     end
 
@@ -69,12 +73,12 @@ module Raif
     end
 
     config.after_initialize do
-      next unless Raif.config.aws_bedrock_titan_embedding_models_enabled
+      next unless Raif.config.bedrock_embedding_models_enabled
 
       require "aws-sdk-bedrockruntime"
 
-      Raif.default_embedding_models[Raif::EmbeddingModels::BedrockTitan].each do |embedding_model_config|
-        Raif.register_embedding_model(Raif::EmbeddingModels::BedrockTitan, **embedding_model_config)
+      Raif.default_embedding_models[Raif::EmbeddingModels::Bedrock].each do |embedding_model_config|
+        Raif.register_embedding_model(Raif::EmbeddingModels::Bedrock, **embedding_model_config)
       end
     end
 
@@ -84,7 +88,7 @@ module Raif
       Raif.config.conversation_types += ["Raif::TestConversation"]
 
       require "#{Raif::Engine.root}/spec/support/test_llm"
-      Raif.register_llm(Raif::Llms::Test, key: :raif_test_llm, api_name: "raif-test-llm")
+      Raif.register_llm(Raif::Llms::TestLlm, key: :raif_test_llm, api_name: "raif-test-llm")
 
       require "#{Raif::Engine.root}/spec/support/test_embedding_model"
       Raif.register_embedding_model(
@@ -99,6 +103,18 @@ module Raif
       Raif.config.validate!
     end
 
+    config.after_initialize do
+      # Check to see if the host app is missing any of our migrations
+      # and print a warning if they are
+      next unless Rails.env.development?
+      next if File.basename($PROGRAM_NAME) == "rake"
+
+      # Skip if we're running inside the engine's own dummy app
+      next if Rails.root.to_s.include?("raif/spec/dummy")
+
+      Raif::MigrationChecker.check_and_warn!
+    end
+
     initializer "raif.assets" do
       if Rails.application.config.respond_to?(:assets)
         Rails.application.config.assets.precompile += [

data/lib/raif/errors/streaming_error.rb

@@ -0,0 +1,18 @@
+# frozen_string_literal: true
+
+module Raif
+  module Errors
+    class StreamingError < StandardError
+      attr_reader :message, :type, :code, :event
+
+      def initialize(message:, type:, event:, code: nil)
+        super
+
+        @message = message
+        @type = type
+        @code = code
+        @event = event
+      end
+    end
+  end
+end

data/lib/raif/errors.rb

@@ -8,3 +8,4 @@ require "raif/errors/open_ai/json_schema_error"
 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"

data/lib/raif/llm_registry.rb

@@ -1,8 +1,8 @@
 # frozen_string_literal: true
 
 module Raif
-  class << self
-    attr_accessor :llm_registry
+  def self.llm_registry
+    @llm_registry ||= {}
   end
 
   def self.register_llm(llm_class, llm_config)
@@ -40,81 +40,170 @@ module Raif
   end
 
   def self.default_llms
+    open_ai_models = [
+      {
+        key: :open_ai_gpt_4o_mini,
+        api_name: "gpt-4o-mini",
+        input_token_cost: 0.15 / 1_000_000,
+        output_token_cost: 0.6 / 1_000_000,
+      },
+      {
+        key: :open_ai_gpt_4o,
+        api_name: "gpt-4o",
+        input_token_cost: 2.5 / 1_000_000,
+        output_token_cost: 10.0 / 1_000_000,
+      },
+      {
+        key: :open_ai_gpt_3_5_turbo,
+        api_name: "gpt-3.5-turbo",
+        input_token_cost: 0.5 / 1_000_000,
+        output_token_cost: 1.5 / 1_000_000,
+        model_provider_settings: { supports_structured_outputs: false }
+      },
+      {
+        key: :open_ai_gpt_4_1,
+        api_name: "gpt-4.1",
+        input_token_cost: 2.0 / 1_000_000,
+        output_token_cost: 8.0 / 1_000_000,
+      },
+      {
+        key: :open_ai_gpt_4_1_mini,
+        api_name: "gpt-4.1-mini",
+        input_token_cost: 0.4 / 1_000_000,
+        output_token_cost: 1.6 / 1_000_000,
+      },
+      {
+        key: :open_ai_gpt_4_1_nano,
+        api_name: "gpt-4.1-nano",
+        input_token_cost: 0.1 / 1_000_000,
+        output_token_cost: 0.4 / 1_000_000,
+      },
+      {
+        key: :open_ai_o1,
+        api_name: "o1",
+        input_token_cost: 15.0 / 1_000_000,
+        output_token_cost: 60.0 / 1_000_000,
+        model_provider_settings: { supports_temperature: false },
+      },
+      {
+        key: :open_ai_o1_mini,
+        api_name: "o1-mini",
+        input_token_cost: 1.5 / 1_000_000,
+        output_token_cost: 6.0 / 1_000_000,
+        model_provider_settings: { supports_temperature: false },
+      },
+      {
+        key: :open_ai_o3,
+        api_name: "o3",
+        input_token_cost: 2.0 / 1_000_000,
+        output_token_cost: 8.0 / 1_000_000,
+        model_provider_settings: { supports_temperature: false },
+      },
+      {
+        key: :open_ai_o3_mini,
+        api_name: "o3-mini",
+        input_token_cost: 1.1 / 1_000_000,
+        output_token_cost: 4.4 / 1_000_000,
+        model_provider_settings: { supports_temperature: false },
+      },
+      {
+        key: :open_ai_o4_mini,
+        api_name: "o4-mini",
+        input_token_cost: 1.1 / 1_000_000,
+        output_token_cost: 4.4 / 1_000_000,
+        model_provider_settings: { supports_temperature: false },
+      },
+    ]
+
+    open_ai_responses_models = open_ai_models.dup.map.with_index do |model, _index|
+      model.merge(
+        key: model[:key].to_s.gsub("open_ai_", "open_ai_responses_").to_sym,
+        supported_provider_managed_tools: [
+          Raif::ModelTools::ProviderManaged::WebSearch,
+          Raif::ModelTools::ProviderManaged::CodeExecution,
+          Raif::ModelTools::ProviderManaged::ImageGeneration
+        ]
+      )
+    end
+
+    # o1-mini is not supported by the OpenAI Responses API.
+    open_ai_responses_models.delete_if{|model| model[:key] == :open_ai_o1_mini }
+
+    # o1-pro and o3-pro are not supported by the OpenAI Completions API, but it is supported by the OpenAI Responses API.
+    open_ai_responses_models << {
+      key: :open_ai_responses_o1_pro,
+      api_name: "o1-pro",
+      input_token_cost: 150.0 / 1_000_000,
+      output_token_cost: 600.0 / 1_000_000,
+      model_provider_settings: { supports_temperature: false },
+    }
+
+    open_ai_responses_models << {
+      key: :open_ai_responses_o3_pro,
+      api_name: "o3-pro",
+      input_token_cost: 20.0 / 1_000_000,
+      output_token_cost: 80.0 / 1_000_000,
+      model_provider_settings: { supports_temperature: false },
+    }
+
     {
-      Raif::Llms::OpenAi => [
-        {
-          key: :open_ai_gpt_4o_mini,
-          api_name: "gpt-4o-mini",
-          input_token_cost: 0.15 / 1_000_000,
-          output_token_cost: 0.6 / 1_000_000,
-        },
-        {
-          key: :open_ai_gpt_4o,
-          api_name: "gpt-4o",
-          input_token_cost: 2.5 / 1_000_000,
-          output_token_cost: 10.0 / 1_000_000,
-        },
-        {
-          key: :open_ai_gpt_3_5_turbo,
-          api_name: "gpt-3.5-turbo",
-          input_token_cost: 0.5 / 1_000_000,
-          output_token_cost: 1.5 / 1_000_000,
-          model_provider_settings: { supports_structured_outputs: false }
-        },
-        {
-          key: :open_ai_gpt_4_1,
-          api_name: "gpt-4.1",
-          input_token_cost: 2.0 / 1_000_000,
-          output_token_cost: 8.0 / 1_000_000,
-        },
-        {
-          key: :open_ai_gpt_4_1_mini,
-          api_name: "gpt-4.1-mini",
-          input_token_cost: 0.4 / 1_000_000,
-          output_token_cost: 1.6 / 1_000_000,
-        },
-        {
-          key: :open_ai_gpt_4_1_nano,
-          api_name: "gpt-4.1-nano",
-          input_token_cost: 0.1 / 1_000_000,
-          output_token_cost: 0.4 / 1_000_000,
-        },
-      ],
+      Raif::Llms::OpenAiCompletions => open_ai_models,
+      Raif::Llms::OpenAiResponses => open_ai_responses_models,
       Raif::Llms::Anthropic => [
         {
           key: :anthropic_claude_4_sonnet,
           api_name: "claude-sonnet-4-20250514",
           input_token_cost: 3.0 / 1_000_000,
           output_token_cost: 15.0 / 1_000_000,
-          max_completion_tokens: 8192
+          max_completion_tokens: 8192,
+          supported_provider_managed_tools: [
+            Raif::ModelTools::ProviderManaged::WebSearch,
+            Raif::ModelTools::ProviderManaged::CodeExecution
+          ]
         },
         {
           key: :anthropic_claude_4_opus,
           api_name: "claude-opus-4-20250514",
           input_token_cost: 15.0 / 1_000_000,
           output_token_cost: 75.0 / 1_000_000,
-          max_completion_tokens: 8192
+          max_completion_tokens: 8192,
+          supported_provider_managed_tools: [
+            Raif::ModelTools::ProviderManaged::WebSearch,
+            Raif::ModelTools::ProviderManaged::CodeExecution
+          ]
         },
         {
           key: :anthropic_claude_3_7_sonnet,
           api_name: "claude-3-7-sonnet-latest",
           input_token_cost: 3.0 / 1_000_000,
           output_token_cost: 15.0 / 1_000_000,
-          max_completion_tokens: 8192
+          max_completion_tokens: 8192,
+          supported_provider_managed_tools: [
+            Raif::ModelTools::ProviderManaged::WebSearch,
+            Raif::ModelTools::ProviderManaged::CodeExecution
+          ]
         },
         {
           key: :anthropic_claude_3_5_sonnet,
           api_name: "claude-3-5-sonnet-latest",
           input_token_cost: 3.0 / 1_000_000,
           output_token_cost: 15.0 / 1_000_000,
-          max_completion_tokens: 8192
+          max_completion_tokens: 8192,
+          supported_provider_managed_tools: [
+            Raif::ModelTools::ProviderManaged::WebSearch,
+            Raif::ModelTools::ProviderManaged::CodeExecution
+          ]
         },
         {
           key: :anthropic_claude_3_5_haiku,
           api_name: "claude-3-5-haiku-latest",
           input_token_cost: 0.8 / 1_000_000,
           output_token_cost: 4.0 / 1_000_000,
-          max_completion_tokens: 8192
+          max_completion_tokens: 8192,
+          supported_provider_managed_tools: [
+            Raif::ModelTools::ProviderManaged::WebSearch,
+            Raif::ModelTools::ProviderManaged::CodeExecution
+          ]
         },
         {
           key: :anthropic_claude_3_opus,
@@ -124,7 +213,7 @@ module Raif
           max_completion_tokens: 4096
         },
       ],
-      Raif::Llms::BedrockClaude => [
+      Raif::Llms::Bedrock => [
         {
           key: :bedrock_claude_4_sonnet,
           api_name: "anthropic.claude-sonnet-4-20250514-v1:0",
@@ -167,6 +256,27 @@ module Raif
           output_token_cost: 0.075 / 1000,
           max_completion_tokens: 4096
         },
+        {
+          key: :bedrock_amazon_nova_micro,
+          api_name: "amazon.nova-micro-v1:0",
+          input_token_cost: 0.0000115 / 1000,
+          output_token_cost: 0.000184 / 1000,
+          max_completion_tokens: 4096
+        },
+        {
+          key: :bedrock_amazon_nova_lite,
+          api_name: "amazon.nova-lite-v1:0",
+          input_token_cost: 0.0000195 / 1000,
+          output_token_cost: 0.000312 / 1000,
+          max_completion_tokens: 4096
+        },
+        {
+          key: :bedrock_amazon_nova_pro,
+          api_name: "amazon.nova-pro-v1:0",
+          input_token_cost: 0.0002625 / 1000,
+          output_token_cost: 0.0042 / 1000,
+          max_completion_tokens: 4096
+        }
       ],
       Raif::Llms::OpenRouter => [
         {
@@ -187,6 +297,18 @@ module Raif
           input_token_cost: 0.02 / 1_000_000,
           output_token_cost: 0.03 / 1_000_000,
         },
+        {
+          key: :open_router_llama_4_maverick,
+          api_name: "meta-llama/llama-4-maverick",
+          input_token_cost: 0.15 / 1_000_000,
+          output_token_cost: 0.60 / 1_000_000,
+        },
+        {
+          key: :open_router_llama_4_scout,
+          api_name: "meta-llama/llama-4-scout",
+          input_token_cost: 0.08 / 1_000_000,
+          output_token_cost: 0.30 / 1_000_000,
+        },
         {
           key: :open_router_gemini_2_0_flash,
           api_name: "google/gemini-2.0-flash-001",

data/lib/raif/migration_checker.rb

@@ -0,0 +1,74 @@
+# frozen_string_literal: true
+
+module Raif
+  module MigrationChecker
+    class << self
+      def uninstalled_migrations
+        engine_migration_names = engine_migration_names_from_context
+        ran_migration_names = ran_migration_names_from_host
+
+        engine_migration_names - ran_migration_names
+      end
+
+      def check_and_warn!
+        return unless defined?(Rails) && Rails.application
+
+        uninstalled = uninstalled_migrations
+        return if uninstalled.empty?
+
+        warning_message = build_warning_message(uninstalled)
+
+        # Output to both logger and STDOUT to ensure visibility
+        Rails.logger&.warn(warning_message)
+        warn warning_message
+      end
+
+    private
+
+      def engine_migration_names_from_context
+        engine_paths = Raif::Engine.paths["db/migrate"].existent
+        return [] if engine_paths.empty?
+
+        ActiveRecord::MigrationContext.new(engine_paths).migrations.map(&:name)
+      rescue => e
+        Rails.logger&.debug("Raif: Could not load engine migrations: #{e.message}")
+        []
+      end
+
+      def ran_migration_names_from_host
+        return [] unless defined?(Rails) && Rails.application
+
+        app_paths = Rails.application.paths["db/migrate"].expanded
+        return [] if app_paths.empty?
+
+        ctx = ActiveRecord::MigrationContext.new(app_paths)
+        ran_versions = ctx.get_all_versions
+        ctx.migrations.select{|m| ran_versions.include?(m.version) }.map(&:name)
+      rescue ActiveRecord::NoDatabaseError
+        # Database doesn't exist yet, so no migrations have been run
+        []
+      rescue => e
+        Rails.logger&.debug("Raif: Could not load migration status: #{e.message}")
+        []
+      end
+
+      def build_warning_message(uninstalled_migration_names)
+        <<~WARNING
+          \e[33m
+          ⚠️  RAIF MIGRATION WARNING ⚠️
+
+          The following Raif migrations have not been run in your application:
+
+          #{uninstalled_migration_names.map { |name| "  • #{name}" }.join("\n")}
+
+          To install and run these migrations:
+
+            rails raif:install:migrations
+            rails db:migrate
+
+          \e[0m
+        WARNING
+      end
+    end
+  end
+end