activeagent 0.5.0rc3 → 0.6.0rc1

data/lib/active_agent/generation_provider/open_router_provider.rb

@@ -4,11 +4,339 @@ require_relative "open_ai_provider"
 module ActiveAgent
   module GenerationProvider
     class OpenRouterProvider < OpenAIProvider
+      # Vision-capable models on OpenRouter
+      VISION_MODELS = [
+        "openai/gpt-4-vision-preview",
+        "openai/gpt-4o",
+        "openai/gpt-4o-mini",
+        "anthropic/claude-3-5-sonnet",
+        "anthropic/claude-3-opus",
+        "anthropic/claude-3-sonnet",
+        "anthropic/claude-3-haiku",
+        "google/gemini-pro-1.5",
+        "google/gemini-pro-vision"
+      ].freeze
+
+      # Models that support structured output
+      STRUCTURED_OUTPUT_MODELS = [
+        "openai/gpt-4o",
+        "openai/gpt-4o-2024-08-06",
+        "openai/gpt-4o-mini",
+        "openai/gpt-4o-mini-2024-07-18",
+        "openai/gpt-4-turbo",
+        "openai/gpt-4-turbo-2024-04-09",
+        "openai/gpt-3.5-turbo-0125",
+        "openai/gpt-3.5-turbo-1106"
+      ].freeze
+
       def initialize(config)
         @config = config
-        @access_token ||= config["api_key"] || config["access_token"] || ENV["OPENROUTER_API_KEY"] || ENV["OPENROUTER_ACCESS_TOKEN"]
+        @access_token = config["api_key"] || config["access_token"] ||
+                       ENV["OPENROUTER_API_KEY"] || ENV["OPENROUTER_ACCESS_TOKEN"]
         @model_name = config["model"]
-        @client = OpenAI::Client.new(uri_base: "https://openrouter.ai/api/v1", access_token: @access_token, log_errors: true)
+
+        # OpenRouter-specific configuration
+        @app_name = config["app_name"] || default_app_name
+        @site_url = config["site_url"] || default_site_url
+        @enable_fallbacks = config["enable_fallbacks"] != false
+        @fallback_models = config["fallback_models"] || []
+        @transforms = config["transforms"] || []
+        @provider_preferences = config["provider"] || {}
+        @track_costs = config["track_costs"] != false
+        @route = config["route"] || "fallback"
+
+        # Data collection preference (allow, deny, or specific provider list)
+        @data_collection = config["data_collection"] || @provider_preferences["data_collection"] || "allow"
+
+        # Initialize OpenAI client with OpenRouter base URL
+        @client = OpenAI::Client.new(
+          uri_base: "https://openrouter.ai/api/v1",
+          access_token: @access_token,
+          log_errors: true,
+          default_headers: openrouter_headers
+        )
+      end
+
+      def generate(prompt)
+        @prompt = prompt
+
+        with_error_handling do
+          parameters = build_openrouter_parameters
+          response = execute_with_fallback(parameters)
+          process_openrouter_response(response)
+        end
+      rescue => e
+        handle_openrouter_error(e)
+      end
+
+      # Helper methods for checking model capabilities
+      def supports_vision?(model = @model_name)
+        VISION_MODELS.include?(model)
+      end
+
+      def supports_structured_output?(model = @model_name)
+        STRUCTURED_OUTPUT_MODELS.include?(model)
+      end
+
+      protected
+
+      def build_provider_parameters
+        # Start with base OpenAI parameters
+        params = super
+
+        # Add OpenRouter-specific parameters
+        add_openrouter_params(params)
+      end
+
+      private
+
+      def default_app_name
+        if defined?(Rails) && Rails.application
+          Rails.application.class.name.split("::").first
+        else
+          "ActiveAgent"
+        end
+      end
+
+      def default_site_url
+        # First check ActiveAgent config
+        return config["default_url_options"]["host"] if config.dig("default_url_options", "host")
+
+        # Then check Rails routes default_url_options
+        if defined?(Rails) && Rails.application&.routes&.default_url_options&.any?
+          host = Rails.application.routes.default_url_options[:host]
+          port = Rails.application.routes.default_url_options[:port]
+          protocol = Rails.application.routes.default_url_options[:protocol] || "https"
+
+          if host
+            url = "#{protocol}://#{host}"
+            url += ":#{port}" if port && port != 80 && port != 443
+            return url
+          end
+        end
+
+        # Finally check ActionMailer config as fallback
+        if defined?(Rails) && Rails.application&.config&.action_mailer&.default_url_options
+          options = Rails.application.config.action_mailer.default_url_options
+          host = options[:host]
+          port = options[:port]
+          protocol = options[:protocol] || "https"
+
+          if host
+            url = "#{protocol}://#{host}"
+            url += ":#{port}" if port && port != 80 && port != 443
+            return url
+          end
+        end
+
+        nil
+      end
+
+      def openrouter_headers
+        headers = {}
+        headers["HTTP-Referer"] = @site_url if @site_url
+        headers["X-Title"] = @app_name if @app_name
+        headers
+      end
+
+      def build_openrouter_parameters
+        parameters = prompt_parameters
+
+        # Handle multiple models for fallback
+        if @fallback_models.present?
+          parameters[:models] = [ @model_name ] + @fallback_models
+          parameters[:route] = @route
+        end
+
+        # Add transforms if specified
+        parameters[:transforms] = @transforms if @transforms.present?
+
+        # Add provider preferences (always include if we have data_collection or other settings)
+        # Check both configured and runtime data_collection values
+        runtime_data_collection = prompt&.options&.key?(:data_collection)
+        if @provider_preferences.present? || @data_collection != "allow" || runtime_data_collection
+          parameters[:provider] = build_provider_preferences
+        end
+
+        parameters
+      end
+
+      def build_provider_preferences
+        prefs = {}
+        prefs[:order] = @provider_preferences["order"] if @provider_preferences["order"]
+        prefs[:require_parameters] = @provider_preferences["require_parameters"] if @provider_preferences.key?("require_parameters")
+        prefs[:allow_fallbacks] = @enable_fallbacks
+
+        # Data collection can be:
+        # - "allow" (default): Allow all providers to collect data
+        # - "deny": Deny all providers from collecting data
+        # - Array of provider names: Only allow these providers to collect data
+        # Check prompt options first (runtime override), then fall back to configured value
+        data_collection = prompt.options[:data_collection] if prompt&.options&.key?(:data_collection)
+        data_collection ||= @data_collection
+        prefs[:data_collection] = data_collection
+
+        prefs.compact
+      end
+
+      def add_openrouter_params(params)
+        # Add OpenRouter-specific routing parameters
+        if @enable_fallbacks && @fallback_models.present?
+          params[:models] = [ @model_name ] + @fallback_models
+          params[:route] = @route
+        end
+
+        # Add transforms
+        params[:transforms] = @transforms if @transforms.present?
+
+        # Add provider configuration (always include if we have data_collection or other settings)
+        # Check both configured and runtime data_collection values
+        runtime_data_collection = prompt&.options&.key?(:data_collection)
+        if @provider_preferences.present? || @data_collection != "allow" || runtime_data_collection
+          params[:provider] = build_provider_preferences
+        end
+
+        # Add plugins (e.g., for PDF processing)
+        if prompt.options[:plugins].present?
+          params[:plugins] = prompt.options[:plugins]
+        end
+
+        params
+      end
+
+      def execute_with_fallback(parameters)
+        parameters[:stream] = provider_stream if prompt.options[:stream] || config["stream"]
+
+        response = @client.chat(parameters: parameters)
+
+        # Log if fallback was used
+        if response.respond_to?(:headers) && response.headers["x-model"] != @model_name
+          Rails.logger.info "[OpenRouter] Fallback model used: #{response.headers['x-model']}" if defined?(Rails)
+        end
+
+        response
+      end
+
+      def process_openrouter_response(response)
+        # Process as normal OpenAI response first
+        if prompt.options[:stream]
+          return @response
+        end
+
+        # Extract standard response
+        message_json = response.dig("choices", 0, "message")
+        message_json["id"] = response.dig("id") if message_json && message_json["id"].blank?
+        message = handle_message(message_json) if message_json
+
+        update_context(prompt: prompt, message: message, response: response) if message
+
+        # Create response with OpenRouter metadata
+        @response = ActiveAgent::GenerationProvider::Response.new(
+          prompt: prompt,
+          message: message,
+          raw_response: response
+        )
+
+        # OpenRouter includes provider and model info directly in the response body
+        if response["provider"] || response["model"]
+          @response.metadata = {
+            provider: response["provider"],
+            model_used: response["model"],
+            fallback_used: response["model"] != @model_name
+          }.compact
+        end
+
+        # Track costs if enabled
+        track_usage(response) if @track_costs && response["usage"]
+
+        @response
+      end
+
+      def add_openrouter_metadata(response, headers)
+        return unless response.respond_to?(:metadata=)
+
+        response.metadata = {
+          provider: headers["x-provider"],
+          model_used: headers["x-model"],
+          trace_id: headers["x-trace-id"],
+          fallback_used: headers["x-model"] != @model_name,
+          ratelimit: {
+            requests_limit: headers["x-ratelimit-requests-limit"],
+            requests_remaining: headers["x-ratelimit-requests-remaining"],
+            requests_reset: headers["x-ratelimit-requests-reset"],
+            tokens_limit: headers["x-ratelimit-tokens-limit"],
+            tokens_remaining: headers["x-ratelimit-tokens-remaining"],
+            tokens_reset: headers["x-ratelimit-tokens-reset"]
+          }.compact
+        }.compact
+      end
+
+      def track_usage(response)
+        return nil unless @track_costs
+        return nil unless response["usage"]
+
+        usage = response["usage"]
+        model = response.dig("model") || @model_name
+
+        # Calculate costs (simplified - would need actual pricing data)
+        cost_info = {
+          model: model,
+          prompt_tokens: usage["prompt_tokens"],
+          completion_tokens: usage["completion_tokens"],
+          total_tokens: usage["total_tokens"]
+        }
+
+        # Log usage information
+        if defined?(Rails)
+          Rails.logger.info "[OpenRouter] Usage: #{cost_info.to_json}"
+
+          # Store in cache if available
+          if Rails.cache
+            cache_key = "openrouter:usage:#{Date.current}"
+            Rails.cache.increment("#{cache_key}:tokens", usage["total_tokens"])
+            Rails.cache.increment("#{cache_key}:requests")
+          end
+        end
+
+        cost_info
+      end
+
+      def handle_openrouter_error(error)
+        error_message = error.message || error.to_s
+
+        case error_message
+        when /rate limit/i
+          handle_rate_limit_error(error)
+        when /insufficient credits|payment required/i
+          handle_insufficient_credits(error)
+        when /no available provider/i
+          handle_no_provider_error(error)
+        when /timeout/i
+          handle_timeout_error(error)
+        else
+          # Fall back to parent error handling
+          super(error) if defined?(super)
+        end
+      end
+
+      def handle_rate_limit_error(error)
+        Rails.logger.error "[OpenRouter] Rate limit exceeded: #{error.message}" if defined?(Rails)
+        raise GenerationProviderError, "OpenRouter rate limit exceeded. Please retry later."
+      end
+
+      def handle_insufficient_credits(error)
+        Rails.logger.error "[OpenRouter] Insufficient credits: #{error.message}" if defined?(Rails)
+        raise GenerationProviderError, "OpenRouter account has insufficient credits."
+      end
+
+      def handle_no_provider_error(error)
+        Rails.logger.error "[OpenRouter] No available provider: #{error.message}" if defined?(Rails)
+        raise GenerationProviderError, "No available provider for the requested model."
+      end
+
+      def handle_timeout_error(error)
+        Rails.logger.error "[OpenRouter] Request timeout: #{error.message}" if defined?(Rails)
+        raise GenerationProviderError, "OpenRouter request timed out."
       end
     end
   end

data/lib/active_agent/generation_provider/parameter_builder.rb

@@ -0,0 +1,119 @@
+# frozen_string_literal: true
+
+module ActiveAgent
+  module GenerationProvider
+    module ParameterBuilder
+      extend ActiveSupport::Concern
+
+      def prompt_parameters(overrides = {})
+        base_params = build_base_parameters
+        provider_params = build_provider_parameters
+
+        # Merge parameters with proper precedence:
+        # 1. Overrides (highest priority)
+        # 2. Prompt options
+        # 3. Provider-specific parameters
+        # 4. Base parameters (lowest priority)
+        base_params
+          .merge(provider_params)
+          .merge(extract_prompt_options)
+          .merge(overrides)
+          .compact
+      end
+
+      protected
+
+      def build_base_parameters
+        {
+          model: determine_model,
+          messages: provider_messages(@prompt.messages),
+          temperature: determine_temperature
+        }.tap do |params|
+          # Add optional parameters if present
+          params[:max_tokens] = determine_max_tokens if determine_max_tokens
+          params[:tools] = format_tools(@prompt.actions) if @prompt.actions.present?
+        end
+      end
+
+      def build_provider_parameters
+        # Override in provider for specific parameters
+        # For example, Anthropic needs 'system' parameter instead of system message
+        {}
+      end
+
+      def extract_prompt_options
+        # Extract relevant options from prompt
+        options = {}
+
+        # Common options that map directly
+        [ :stream, :top_p, :frequency_penalty, :presence_penalty, :seed, :stop, :user ].each do |key|
+          options[key] = @prompt.options[key] if @prompt.options.key?(key)
+        end
+
+        # Handle response format for structured output
+        if @prompt.output_schema.present?
+          options[:response_format] = build_response_format
+        end
+
+        options
+      end
+
+      def determine_model
+        @prompt.options[:model] || @model_name || @config["model"]
+      end
+
+      def determine_temperature
+        @prompt.options[:temperature] || @config["temperature"] || 0.7
+      end
+
+      def determine_max_tokens
+        @prompt.options[:max_tokens] || @config["max_tokens"]
+      end
+
+      def build_response_format
+        # Default OpenAI-style response format
+        # Override in provider for different formats
+        {
+          type: "json_schema",
+          json_schema: @prompt.output_schema
+        }
+      end
+
+      # Embedding-specific parameters
+      def embeddings_parameters(input: nil, model: nil, **options)
+        {
+          model: model || determine_embedding_model,
+          input: input || format_embedding_input,
+          dimensions: options[:dimensions] || @config["embedding_dimensions"],
+          encoding_format: options[:encoding_format] || "float"
+        }.compact
+      end
+
+      def determine_embedding_model
+        @prompt.options[:embedding_model] || @config["embedding_model"] || "text-embedding-3-small"
+      end
+
+      def format_embedding_input
+        # Handle single or batch embedding inputs
+        if @prompt.message
+          @prompt.message.content
+        elsif @prompt.messages
+          @prompt.messages.map(&:content)
+        else
+          nil
+        end
+      end
+
+      module ClassMethods
+        # Class-level configuration for default parameters
+        def default_parameters(params = {})
+          @default_parameters = params
+        end
+
+        def get_default_parameters
+          @default_parameters || {}
+        end
+      end
+    end
+  end
+end

data/lib/active_agent/generation_provider/response.rb

@@ -4,11 +4,13 @@ module ActiveAgent
   module GenerationProvider
     class Response
       attr_reader :message, :prompt, :raw_response
+      attr_accessor :metadata
 
-      def initialize(prompt:, message: nil, raw_response: nil)
+      def initialize(prompt:, message: nil, raw_response: nil, metadata: nil)
         @prompt = prompt
         @message = message || prompt.message
         @raw_response = raw_response
+        @metadata = metadata || {}
       end
 
       # Extract usage statistics from the raw response

data/lib/active_agent/generation_provider/stream_processing.rb

@@ -0,0 +1,58 @@
+# frozen_string_literal: true
+
+module ActiveAgent
+  module GenerationProvider
+    module StreamProcessing
+      extend ActiveSupport::Concern
+
+      included do
+        attr_accessor :stream_buffer, :stream_context
+      end
+
+      def provider_stream
+        agent_stream = prompt.options[:stream]
+        message = initialize_stream_message
+
+        @response = ActiveAgent::GenerationProvider::Response.new(prompt:, message:)
+
+        proc do |chunk|
+          process_stream_chunk(chunk, message, agent_stream)
+        end
+      end
+
+      protected
+
+      def initialize_stream_message
+        ActiveAgent::ActionPrompt::Message.new(content: "", role: :assistant)
+      end
+
+      def process_stream_chunk(chunk, message, agent_stream)
+        # Default implementation - must be overridden in provider
+        raise NotImplementedError, "Providers must implement process_stream_chunk"
+      end
+
+      def handle_stream_delta(delta, message, agent_stream)
+        # Common delta handling logic
+        new_content = extract_content_from_delta(delta)
+        if new_content && !new_content.blank?
+          message.content += new_content
+          agent_stream&.call(message, new_content, false, prompt.action_name)
+        end
+      end
+
+      def extract_content_from_delta(delta)
+        # Default extraction - override if needed
+        delta if delta.is_a?(String)
+      end
+
+      def handle_tool_stream_chunk(chunk, message, agent_stream)
+        # Handle tool calls in streaming
+        # Override in provider for specific implementation
+      end
+
+      def finalize_stream(message, agent_stream)
+        agent_stream&.call(message, nil, true, prompt.action_name)
+      end
+    end
+  end
+end

data/lib/active_agent/generation_provider/tool_management.rb

@@ -0,0 +1,142 @@
+# frozen_string_literal: true
+
+module ActiveAgent
+  module GenerationProvider
+    module ToolManagement
+      extend ActiveSupport::Concern
+
+      def format_tools(tools)
+        return nil if tools.blank?
+
+        tools.map do |tool|
+          format_single_tool(tool)
+        end
+      end
+
+      def handle_actions(tool_calls)
+        return [] if tool_calls.nil? || tool_calls.empty?
+
+        tool_calls.map do |tool_call|
+          parse_tool_call(tool_call)
+        end.compact
+      end
+
+      protected
+
+      def format_single_tool(tool)
+        # Default implementation for OpenAI-style tools
+        if tool["function"] || tool[:function]
+          # Tool already has the correct structure
+          tool
+        else
+          # Legacy format - wrap in function structure
+          wrap_tool_in_function(tool)
+        end
+      end
+
+      def wrap_tool_in_function(tool)
+        {
+          type: "function",
+          function: {
+            name: tool["name"] || tool[:name],
+            description: tool["description"] || tool[:description],
+            parameters: tool["parameters"] || tool[:parameters]
+          }
+        }
+      end
+
+      def parse_tool_call(tool_call)
+        # Skip if no function information
+        return nil if tool_call.nil?
+
+        # Extract tool information based on provider format
+        tool_id = extract_tool_id(tool_call)
+        tool_name = extract_tool_name(tool_call)
+        tool_params = extract_tool_params(tool_call)
+
+        # Skip if no name found
+        return nil if tool_name.blank?
+
+        ActiveAgent::ActionPrompt::Action.new(
+          id: tool_id,
+          name: tool_name,
+          params: tool_params
+        )
+      end
+
+      def extract_tool_id(tool_call)
+        tool_call["id"] || tool_call[:id]
+      end
+
+      def extract_tool_name(tool_call)
+        # Try different paths for tool name
+        tool_call.dig("function", "name") ||
+          tool_call.dig(:function, :name) ||
+          tool_call["name"] ||
+          tool_call[:name]
+      end
+
+      def extract_tool_params(tool_call)
+        # Try different paths for tool parameters/arguments
+        args = tool_call.dig("function", "arguments") ||
+               tool_call.dig(:function, :arguments) ||
+               tool_call["arguments"] ||
+               tool_call[:arguments] ||
+               tool_call["input"] ||
+               tool_call[:input]
+
+        return nil if args.blank?
+
+        # Parse JSON string if needed
+        if args.is_a?(String)
+          begin
+            JSON.parse(args, symbolize_names: true)
+          rescue JSON::ParserError
+            nil
+          end
+        else
+          args
+        end
+      end
+
+      # Provider-specific tool format methods
+      # Override these in specific providers
+
+      def format_tools_for_anthropic(tools)
+        # Anthropic-specific tool format
+        tools.map do |tool|
+          {
+            name: extract_tool_name_from_schema(tool),
+            description: extract_tool_description_from_schema(tool),
+            input_schema: extract_tool_parameters_from_schema(tool)
+          }
+        end
+      end
+
+      def format_tools_for_openai(tools)
+        # OpenAI-specific tool format (default)
+        format_tools(tools)
+      end
+
+      private
+
+      def extract_tool_name_from_schema(tool)
+        tool["name"] || tool[:name] ||
+        tool.dig("function", "name") || tool.dig(:function, "name") ||
+        tool.dig("function", :name) || tool.dig(:function, :name)
+      end
+
+      def extract_tool_description_from_schema(tool)
+        tool["description"] || tool[:description] ||
+        tool.dig("function", "description") || tool.dig(:function, "description") ||
+        tool.dig("function", :description) || tool.dig(:function, :description)
+      end
+
+      def extract_tool_parameters_from_schema(tool)
+        tool["parameters"] || tool[:parameters] ||
+        tool.dig("function", "parameters") || tool.dig(:function, "parameters") ||
+        tool.dig("function", :parameters) || tool.dig(:function, :parameters)
+      end
+    end
+  end
+end

data/lib/active_agent/generation_provider.rb

@@ -14,7 +14,6 @@ module ActiveAgent
     module ClassMethods
       def configuration(name_or_provider, **options)
         config = ActiveAgent.config[name_or_provider.to_s] || ActiveAgent.config.dig(ENV["RAILS_ENV"], name_or_provider.to_s) || {}
-
         config = { "service" => "OpenAI" } if config.empty? && name_or_provider == :openai
         config.merge!(options)
       raise "Failed to load provider #{name_or_provider}: configuration not found for provider"  if config["service"].nil?
@@ -26,6 +25,7 @@ module ActiveAgent
       def configure_provider(config)
         service_name = config["service"]
         require "active_agent/generation_provider/#{service_name.underscore}_provider"
+
         ActiveAgent::GenerationProvider.const_get("#{service_name.camelize}Provider").new(config)
       end