raif 1.3.0 → 1.4.0

data/lib/raif/llm_registry.rb

@@ -41,6 +41,41 @@ module Raif
 
   def self.default_llms
     open_ai_models = [
+      {
+        key: :open_ai_gpt_5_2,
+        api_name: "gpt-5.2",
+        input_token_cost: 1.75 / 1_000_000,
+        output_token_cost: 14.0 / 1_000_000,
+        model_provider_settings: { supports_temperature: false },
+      },
+      {
+        key: :open_ai_gpt_5_1,
+        api_name: "gpt-5.1",
+        input_token_cost: 1.25 / 1_000_000,
+        output_token_cost: 10.0 / 1_000_000,
+        model_provider_settings: { supports_temperature: false },
+      },
+      {
+        key: :open_ai_gpt_5,
+        api_name: "gpt-5",
+        input_token_cost: 1.25 / 1_000_000,
+        output_token_cost: 10.0 / 1_000_000,
+        model_provider_settings: { supports_temperature: false },
+      },
+      {
+        key: :open_ai_gpt_5_mini,
+        api_name: "gpt-5-mini",
+        input_token_cost: 0.25 / 1_000_000,
+        output_token_cost: 2.0 / 1_000_000,
+        model_provider_settings: { supports_temperature: false },
+      },
+      {
+        key: :open_ai_gpt_5_nano,
+        api_name: "gpt-5-nano",
+        input_token_cost: 0.05 / 1_000_000,
+        output_token_cost: 0.4 / 1_000_000,
+        model_provider_settings: { supports_temperature: false },
+      },
       {
         key: :open_ai_gpt_4o_mini,
         api_name: "gpt-4o-mini",
@@ -113,27 +148,6 @@ module Raif
         output_token_cost: 4.4 / 1_000_000,
         model_provider_settings: { supports_temperature: false },
       },
-      {
-        key: :open_ai_gpt_5,
-        api_name: "gpt-5",
-        input_token_cost: 1.25 / 1_000_000,
-        output_token_cost: 10.0 / 1_000_000,
-        model_provider_settings: { supports_temperature: false },
-      },
-      {
-        key: :open_ai_gpt_5_mini,
-        api_name: "gpt-5-mini",
-        input_token_cost: 0.25 / 1_000_000,
-        output_token_cost: 2.0 / 1_000_000,
-        model_provider_settings: { supports_temperature: false },
-      },
-      {
-        key: :open_ai_gpt_5_nano,
-        api_name: "gpt-5-nano",
-        input_token_cost: 0.05 / 1_000_000,
-        output_token_cost: 0.4 / 1_000_000,
-        model_provider_settings: { supports_temperature: false },
-      }
     ]
 
     open_ai_responses_models = open_ai_models.dup.map.with_index do |model, _index|
@@ -172,11 +186,33 @@ module Raif
       Raif::Llms::OpenAiResponses => open_ai_responses_models,
       Raif::Llms::Anthropic => [
         {
-          key: :anthropic_claude_4_sonnet,
-          api_name: "claude-sonnet-4-20250514",
+          key: :anthropic_claude_4_5_sonnet,
+          api_name: "claude-sonnet-4-5",
           input_token_cost: 3.0 / 1_000_000,
           output_token_cost: 15.0 / 1_000_000,
-          max_completion_tokens: 8192,
+          max_completion_tokens: 64_000,
+          supported_provider_managed_tools: [
+            Raif::ModelTools::ProviderManaged::WebSearch,
+            Raif::ModelTools::ProviderManaged::CodeExecution
+          ]
+        },
+        {
+          key: :anthropic_claude_4_5_haiku,
+          api_name: "claude-haiku-4-5",
+          input_token_cost: 1.0 / 1_000_000,
+          output_token_cost: 5.0 / 1_000_000,
+          max_completion_tokens: 64_000,
+          supported_provider_managed_tools: [
+            Raif::ModelTools::ProviderManaged::WebSearch,
+            Raif::ModelTools::ProviderManaged::CodeExecution
+          ]
+        },
+        {
+          key: :anthropic_claude_4_1_opus,
+          api_name: "claude-opus-4-1",
+          input_token_cost: 15.0 / 1_000_000,
+          output_token_cost: 75.0 / 1_000_000,
+          max_completion_tokens: 32_000,
           supported_provider_managed_tools: [
             Raif::ModelTools::ProviderManaged::WebSearch,
             Raif::ModelTools::ProviderManaged::CodeExecution
@@ -193,6 +229,17 @@ module Raif
             Raif::ModelTools::ProviderManaged::CodeExecution
           ]
         },
+        {
+          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,
+          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",
@@ -235,6 +282,27 @@ module Raif
         },
       ],
       Raif::Llms::Bedrock => [
+        {
+          key: :bedrock_claude_4_5_sonnet,
+          api_name: "anthropic.claude-sonnet-4-5-20250929-v1:0",
+          input_token_cost: 0.003 / 1000,
+          output_token_cost: 0.015 / 1000,
+          max_completion_tokens: 64_000
+        },
+        {
+          key: :bedrock_claude_4_5_haiku,
+          api_name: "anthropic.claude-haiku-4-5-20251001-v1:0",
+          input_token_cost: 0.001 / 1000,
+          output_token_cost: 0.005 / 1000,
+          max_completion_tokens: 64_000
+        },
+        {
+          key: :bedrock_claude_4_1_opus,
+          api_name: "anthropic.claude-opus-4-1-20250805-v1:0",
+          input_token_cost: 0.015 / 1000,
+          output_token_cost: 0.075 / 1000,
+          max_completion_tokens: 32_000
+        },
         {
           key: :bedrock_claude_4_sonnet,
           api_name: "anthropic.claude-sonnet-4-20250514-v1:0",
@@ -250,15 +318,15 @@ module Raif
           max_completion_tokens: 8192
         },
         {
-          key: :bedrock_claude_3_5_sonnet,
-          api_name: "anthropic.claude-3-5-sonnet-20241022-v2:0",
+          key: :bedrock_claude_3_7_sonnet,
+          api_name: "anthropic.claude-3-7-sonnet-20250219-v1:0",
           input_token_cost: 0.003 / 1000,
           output_token_cost: 0.015 / 1000,
           max_completion_tokens: 8192
         },
         {
-          key: :bedrock_claude_3_7_sonnet,
-          api_name: "anthropic.claude-3-7-sonnet-20250219-v1:0",
+          key: :bedrock_claude_3_5_sonnet,
+          api_name: "anthropic.claude-3-5-sonnet-20241022-v2:0",
           input_token_cost: 0.003 / 1000,
           output_token_cost: 0.015 / 1000,
           max_completion_tokens: 8192
@@ -306,6 +374,60 @@ module Raif
           input_token_cost: 3.0 / 1_000_000,
           output_token_cost: 15.0 / 1_000_000,
         },
+        {
+          key: :open_router_deepseek_chat_v3,
+          api_name: "deepseek/deepseek-chat-v3-0324",
+          input_token_cost: 0.27 / 1_000_000,
+          output_token_cost: 1.1 / 1_000_000,
+        },
+        {
+          key: :open_router_deepseek_v3_1,
+          api_name: "deepseek/deepseek-chat-v3.1",
+          input_token_cost: 0.25 / 1_000_000,
+          output_token_cost: 1.0 / 1_000_000,
+        },
+        {
+          key: :open_router_gemini_2_0_flash,
+          api_name: "google/gemini-2.0-flash-001",
+          input_token_cost: 0.1 / 1_000_000,
+          output_token_cost: 0.4 / 1_000_000,
+        },
+        {
+          key: :open_router_gemini_2_5_flash,
+          api_name: "google/gemini-2.5-flash",
+          input_token_cost: 0.3 / 1_000_000,
+          output_token_cost: 2.5 / 1_000_000,
+        },
+        {
+          key: :open_router_gemini_2_5_pro,
+          api_name: "google/gemini-2.5-pro",
+          input_token_cost: 1.25 / 1_000_000,
+          output_token_cost: 10.0 / 1_000_000,
+        },
+        {
+          key: :open_router_gemini_3_pro_preview,
+          api_name: "google/gemini-3-pro-preview",
+          input_token_cost: 2.0 / 1_000_000,
+          output_token_cost: 12.0 / 1_000_000,
+        },
+        {
+          key: :open_router_grok_4,
+          api_name: "x-ai/grok-4",
+          input_token_cost: 3.0 / 1_000_000,
+          output_token_cost: 15.0 / 1_000_000,
+        },
+        {
+          key: :open_router_grok_4_1_fast,
+          api_name: "x-ai/grok-4.1-fast",
+          input_token_cost: 0.2 / 1_000_000,
+          output_token_cost: 0.5 / 1_000_000,
+        },
+        {
+          key: :open_router_kimi_k2_thinking,
+          api_name: "moonshotai/kimi-k2-thinking",
+          input_token_cost: 0.45 / 1_000_000,
+          output_token_cost: 2.35 / 1_000_000,
+        },
         {
           key: :open_router_llama_3_3_70b_instruct,
           api_name: "meta-llama/llama-3.3-70b-instruct",
@@ -331,16 +453,22 @@ module Raif
           output_token_cost: 0.30 / 1_000_000,
         },
         {
-          key: :open_router_gemini_2_0_flash,
-          api_name: "google/gemini-2.0-flash-001",
-          input_token_cost: 0.1 / 1_000_000,
-          output_token_cost: 0.4 / 1_000_000,
+          key: :open_router_minimax_m2,
+          api_name: "minimax/minimax-m2",
+          input_token_cost: 0.255 / 1_000_000,
+          output_token_cost: 1.02 / 1_000_000,
         },
         {
-          key: :open_router_deepseek_chat_v3,
-          api_name: "deepseek/deepseek-chat-v3-0324",
-          input_token_cost: 0.27 / 1_000_000,
-          output_token_cost: 1.1 / 1_000_000,
+          key: :open_router_mistral_large_3_2512,
+          api_name: "mistralai/mistral-large-2512",
+          input_token_cost: 0.50 / 1_000_000,
+          output_token_cost: 1.5 / 1_000_000,
+        },
+        {
+          key: :open_router_mistral_small_3_2_24b,
+          api_name: "mistralai/mistral-small-3.2-24b-instruct",
+          input_token_cost: 0.06 / 1_000_000,
+          output_token_cost: 0.18 / 1_000_000,
         },
         {
           key: :open_router_open_ai_gpt_oss_120b,
@@ -353,7 +481,49 @@ module Raif
           api_name: "gpt-oss-20b",
           input_token_cost: 0.05 / 1_000_000,
           output_token_cost: 0.2 / 1_000_000,
-        }
+        },
+      ],
+      Raif::Llms::Google => [
+        {
+          key: :google_gemini_3_0_pro,
+          api_name: "gemini-3-pro-preview",
+          input_token_cost: 2.0 / 1_000_000,
+          output_token_cost: 12.0 / 1_000_000,
+          supported_provider_managed_tools: [
+            Raif::ModelTools::ProviderManaged::WebSearch,
+            Raif::ModelTools::ProviderManaged::CodeExecution
+          ]
+        },
+        {
+          key: :google_gemini_3_0_flash,
+          api_name: "gemini-3-flash-preview",
+          input_token_cost: 0.5 / 1_000_000,
+          output_token_cost: 3.0 / 1_000_000,
+          supported_provider_managed_tools: [
+            Raif::ModelTools::ProviderManaged::WebSearch,
+            Raif::ModelTools::ProviderManaged::CodeExecution
+          ]
+        },
+        {
+          key: :google_gemini_2_5_pro,
+          api_name: "gemini-2.5-pro-preview-06-05",
+          input_token_cost: 1.25 / 1_000_000,
+          output_token_cost: 10.0 / 1_000_000,
+          supported_provider_managed_tools: [
+            Raif::ModelTools::ProviderManaged::WebSearch,
+            Raif::ModelTools::ProviderManaged::CodeExecution
+          ]
+        },
+        {
+          key: :google_gemini_2_5_flash,
+          api_name: "gemini-2.5-flash",
+          input_token_cost: 0.3 / 1_000_000,
+          output_token_cost: 2.5 / 1_000_000,
+          supported_provider_managed_tools: [
+            Raif::ModelTools::ProviderManaged::WebSearch,
+            Raif::ModelTools::ProviderManaged::CodeExecution
+          ]
+        },
       ]
     }
   end

data/lib/raif/messages.rb

@@ -0,0 +1,180 @@
+# frozen_string_literal: true
+
+module Raif
+  # Message types for agent conversation_history and conversation llm_messages.
+  #
+  # These classes provide a structured API for creating messages that get stored
+  # as JSONB and passed to LLM providers. Each class has:
+  # - Named parameters for initialization
+  # - `to_h` for converting to hash format (for storage/API calls)
+  # - `from_h` class method for deserializing from stored hashes
+  #
+  # @example Creating messages
+  #   message = Raif::Messages::ToolCall.new(
+  #     name: "wikipedia_search",
+  #     arguments: { query: "Ruby" },
+  #     provider_tool_call_id: "call_123"
+  #   )
+  #   conversation_history << message.to_h
+  #
+  # @example Deserializing stored messages
+  #   messages = Raif::Messages.from_array(agent.conversation_history)
+  #   messages.each do |msg|
+  #     case msg
+  #     when Raif::Messages::ToolCall
+  #       puts "Tool: #{msg.name}"
+  #     when Raif::Messages::UserMessage
+  #       puts "User: #{msg.content}"
+  #     end
+  #   end
+  module Messages
+    # User role message
+    class UserMessage
+      attr_reader :content
+
+      # @param content [String] The user's message content
+      def initialize(content:)
+        @content = content
+      end
+
+      # @return [Hash] Hash representation for JSONB storage and LLM APIs
+      def to_h
+        { "role" => "user", "content" => content }
+      end
+
+      # Deserialize from a hash
+      # @param hash [Hash] A hash with "content" key
+      # @return [UserMessage]
+      def self.from_h(hash)
+        new(content: hash["content"])
+      end
+    end
+
+    # Assistant role message
+    class AssistantMessage
+      attr_reader :content
+
+      # @param content [String] The assistant's message content
+      def initialize(content:)
+        @content = content
+      end
+
+      # @return [Hash] Hash representation for JSONB storage and LLM APIs
+      def to_h
+        { "role" => "assistant", "content" => content }
+      end
+
+      # Deserialize from a hash
+      # @param hash [Hash] A hash with "content" key
+      # @return [AssistantMessage]
+      def self.from_h(hash)
+        new(content: hash["content"])
+      end
+    end
+
+    # Tool invocation request from the assistant
+    class ToolCall
+      attr_reader :provider_tool_call_id, :name, :arguments, :assistant_message, :provider_metadata
+
+      # @param name [String] The tool name (snake_case)
+      # @param arguments [Hash] The arguments passed to the tool
+      # @param provider_tool_call_id [String, nil] Provider-assigned ID for the tool call
+      # @param assistant_message [String, nil] Optional assistant message accompanying the tool call
+      # @param provider_metadata [Hash, nil] Provider-specific metadata (e.g., Google's thoughtSignature)
+      def initialize(name:, arguments:, provider_tool_call_id: nil, assistant_message: nil, provider_metadata: nil)
+        @provider_tool_call_id = provider_tool_call_id
+        @name = name
+        @arguments = arguments
+        @assistant_message = assistant_message
+        @provider_metadata = provider_metadata
+      end
+
+      # @return [Hash] Hash representation for JSONB storage and LLM APIs
+      def to_h
+        {
+          "type" => "tool_call",
+          "provider_tool_call_id" => provider_tool_call_id,
+          "name" => name,
+          "arguments" => arguments,
+          "assistant_message" => assistant_message,
+          "provider_metadata" => provider_metadata
+        }.compact
+      end
+
+      # Deserialize from a hash
+      # @param hash [Hash] A hash with tool call fields
+      # @return [ToolCall]
+      def self.from_h(hash)
+        new(
+          name: hash["name"],
+          arguments: hash["arguments"],
+          provider_tool_call_id: hash["provider_tool_call_id"],
+          assistant_message: hash["assistant_message"],
+          provider_metadata: hash["provider_metadata"]
+        )
+      end
+    end
+
+    # Result of a tool invocation
+    class ToolCallResult
+      attr_reader :provider_tool_call_id, :name, :result
+
+      # @param result [Hash, String] The result returned by the tool
+      # @param provider_tool_call_id [String, nil] Provider-assigned ID matching the tool call
+      # @param name [String, nil] The tool name (required by some providers like Google)
+      def initialize(result:, provider_tool_call_id: nil, name: nil)
+        @provider_tool_call_id = provider_tool_call_id
+        @name = name
+        @result = result
+      end
+
+      # @return [Hash] Hash representation for JSONB storage and LLM APIs
+      def to_h
+        {
+          "type" => "tool_call_result",
+          "provider_tool_call_id" => provider_tool_call_id,
+          "name" => name,
+          "result" => result
+        }.compact
+      end
+
+      # Deserialize from a hash
+      # @param hash [Hash] A hash with tool call result fields
+      # @return [ToolCallResult]
+      def self.from_h(hash)
+        new(
+          provider_tool_call_id: hash["provider_tool_call_id"],
+          name: hash["name"],
+          result: hash["result"]
+        )
+      end
+    end
+
+    class << self
+      # Deserialize a single message hash into the appropriate message class
+      # @param hash [Hash] A message hash with either "role" or "type" key
+      # @return [UserMessage, AssistantMessage, ToolCall, ToolCallResult]
+      # @raise [ArgumentError] if the hash doesn't match a known message type
+      def from_h(hash)
+        if hash["type"] == "tool_call"
+          ToolCall.from_h(hash)
+        elsif hash["type"] == "tool_call_result"
+          ToolCallResult.from_h(hash)
+        elsif hash["role"] == "user"
+          UserMessage.from_h(hash)
+        elsif hash["role"] == "assistant"
+          AssistantMessage.from_h(hash)
+        else
+          raise ArgumentError, "Unknown message type: #{hash.inspect}"
+        end
+      end
+
+      # Deserialize an array of message hashes
+      # @param messages [Array<Hash>] Array of message hashes
+      # @return [Array<UserMessage, AssistantMessage, ToolCall, ToolCallResult>]
+      def from_array(messages)
+        messages.map { |msg| from_h(msg) }
+      end
+    end
+  end
+end

data/lib/raif/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Raif
-  VERSION = "1.3.0"
+  VERSION = "1.4.0"
 end

data/lib/raif.rb

@@ -10,6 +10,7 @@ require "raif/llm_registry"
 require "raif/embedding_model_registry"
 require "raif/json_schema_builder"
 require "raif/migration_checker"
+require "raif/messages"
 
 require "faraday"
 require "event_stream_parser"
@@ -41,4 +42,12 @@ module Raif
   def self.running_evals?
     ENV["RAIF_RUNNING_EVALS"] == "true"
   end
+
+  def self.default_request_options
+    {
+      open_timeout: config.request_open_timeout,
+      read_timeout: config.request_read_timeout,
+      write_timeout: config.request_write_timeout
+    }.compact
+  end
 end

data/lib/tasks/annotate_rb.rake

@@ -0,0 +1,10 @@
+# frozen_string_literal: true
+
+# This rake task was added by annotate_rb gem.
+
+# Can set `ANNOTATERB_SKIP_ON_DB_TASKS` to be anything to skip this
+if Rails.env.development? && ENV["ANNOTATERB_SKIP_ON_DB_TASKS"].nil?
+  require "annotate_rb"
+
+  AnnotateRb::Core.load_rake_tasks
+end

data/spec/support/rspec_helpers.rb

@@ -3,11 +3,11 @@
 module Raif
   module RspecHelpers
 
-    def stubbed_llm(llm_model_key, &block)
+    def stubbed_llm(llm_model_key, source_instance, &block)
       test_llm = Raif.llm(llm_model_key.to_sym)
 
       allow(test_llm).to receive(:perform_model_completion!) do |model_completion|
-        result = block.call(model_completion.messages, model_completion)
+        result = block.call(model_completion.messages, model_completion, source_instance)
         model_completion.raw_response = result if result.is_a?(String)
         model_completion.completion_tokens = rand(100..2000)
         model_completion.prompt_tokens = rand(100..2000)
@@ -24,10 +24,10 @@ module Raif
       allow(Raif.config).to receive(:llm_api_requests_enabled){ true }
 
       if task.is_a?(Raif::Task)
-        allow(task).to receive(:llm){ stubbed_llm(task.llm_model_key, &block) }
+        allow(task).to receive(:llm){ stubbed_llm(task.llm_model_key, task, &block) }
       else
         allow_any_instance_of(task).to receive(:llm) do |task_instance|
-          stubbed_llm(task_instance.llm_model_key, &block)
+          stubbed_llm(task_instance.llm_model_key, task_instance, &block)
         end
       end
     end
@@ -36,10 +36,10 @@ module Raif
       allow(Raif.config).to receive(:llm_api_requests_enabled){ true }
 
       if conversation.is_a?(Raif::Conversation)
-        allow(conversation).to receive(:llm){ stubbed_llm(conversation.llm_model_key, &block) }
+        allow(conversation).to receive(:llm){ stubbed_llm(conversation.llm_model_key, conversation, &block) }
       else
         allow_any_instance_of(conversation).to receive(:llm) do |conversation_instance|
-          stubbed_llm(conversation_instance.llm_model_key, &block)
+          stubbed_llm(conversation_instance.llm_model_key, conversation_instance, &block)
         end
       end
     end
@@ -48,10 +48,10 @@ module Raif
       allow(Raif.config).to receive(:llm_api_requests_enabled){ true }
 
       if agent.is_a?(Raif::Agent)
-        allow(agent).to receive(:llm){ stubbed_llm(agent.llm_model_key, &block) }
+        allow(agent).to receive(:llm){ stubbed_llm(agent.llm_model_key, agent, &block) }
       else
         allow_any_instance_of(agent).to receive(:llm) do |agent_instance|
-          stubbed_llm(agent_instance.llm_model_key, &block)
+          stubbed_llm(agent_instance.llm_model_key, agent_instance, &block)
         end
       end
     end