ruby_llm-agents 1.3.3 → 2.0.0

data/lib/ruby_llm/agents/pipeline/middleware/budget.rb

@@ -57,25 +57,51 @@ module RubyLLM
 
           # Checks budget before execution
           #
+          # For tenants, checks budget via counter columns on the tenant model.
+          # For non-tenant usage, falls back to BudgetTracker (cache-based).
+          #
           # @param context [Context] The execution context
           # @raise [BudgetExceededError] If budget exceeded with hard enforcement
           def check_budget!(context)
+            if context.tenant_id.present?
+              tenant = RubyLLM::Agents::Tenant.find_by(tenant_id: context.tenant_id)
+              if tenant
+                tenant.check_budget!(context.agent_class&.name)
+                return
+              end
+            end
+
+            # Fallback to cache-based checking (non-tenant or no tenant record)
             BudgetTracker.check_budget!(
               context.agent_class&.name,
               tenant_id: context.tenant_id
             )
           rescue RubyLLM::Agents::Reliability::BudgetExceededError
-            # Re-raise budget errors
             raise
           rescue StandardError => e
-            # Log but don't fail on budget check errors
             error("Budget check failed: #{e.message}")
           end
 
           # Records spend after execution
           #
+          # For tenants, uses atomic SQL increment via tenant.record_execution!.
+          # For non-tenant usage, falls back to BudgetTracker (cache-based).
+          #
           # @param context [Context] The execution context
           def record_spend!(context)
+            if context.tenant_id.present?
+              tenant = RubyLLM::Agents::Tenant.find_by(tenant_id: context.tenant_id)
+              if tenant
+                tenant.record_execution!(
+                  cost: context.total_cost || 0,
+                  tokens: context.total_tokens || 0,
+                  error: context.error?
+                )
+                return
+              end
+            end
+
+            # Fallback for non-tenant usage
             return unless context.total_cost&.positive?
 
             BudgetTracker.record_spend!(
@@ -84,7 +110,6 @@ module RubyLLM
               tenant_id: context.tenant_id
             )
 
-            # Also record tokens if available
             if context.total_tokens&.positive?
               BudgetTracker.record_tokens!(
                 context.agent_class&.name,
@@ -93,7 +118,6 @@ module RubyLLM
               )
             end
           rescue StandardError => e
-            # Log but don't fail on spend recording errors
             error("Failed to record spend: #{e.message}")
           end
         end

data/lib/ruby_llm/agents/pipeline/middleware/cache.rb

@@ -23,13 +23,6 @@ module RubyLLM
         #     cache_for 1.hour
         #   end
         #
-        # @example Cache versioning
-        #   class MyEmbedder < RubyLLM::Agents::Embedder
-        #     model "text-embedding-3-small"
-        #     version "2.0"  # Change to invalidate cache
-        #     cache_for 1.hour
-        #   end
-        #
         class Cache < Base
           # Process caching
           #
@@ -91,14 +84,15 @@ module RubyLLM
 
           # Generates a cache key for the context
           #
-          # The cache key includes:
+          # Cache keys are content-based, including:
           # - Namespace prefix
           # - Agent type
           # - Agent class name
-          # - Version (for cache invalidation)
           # - Model
           # - SHA256 hash of input
           #
+          # This means caches automatically invalidate when inputs change.
+          #
           # @param context [Context] The execution context
           # @return [String] The cache key
           def generate_cache_key(context)
@@ -106,7 +100,6 @@ module RubyLLM
               "ruby_llm_agents",
               context.agent_type,
               context.agent_class&.name,
-              config(:version, "1.0"),
               context.model,
               hash_input(context.input)
             ].compact

data/lib/ruby_llm/agents/pipeline/middleware/instrumentation.rb

@@ -18,7 +18,6 @@ module RubyLLM
         # Tracking is enabled/disabled per agent type via configuration:
         # - track_executions (conversation agents)
         # - track_embeddings
-        # - track_moderations
         # - track_image_generations
         # - track_audio
         #
@@ -92,7 +91,15 @@ module RubyLLM
             return nil if context.cached? && !track_cache_hits?
 
             data = build_running_execution_data(context)
-            Execution.create!(data)
+            execution = Execution.create!(data)
+
+            # Create detail record with parameters
+            params = sanitize_parameters(context)
+            if params.present? && params != {}
+              execution.create_detail!(parameters: params)
+            end
+
+            execution
           rescue StandardError => e
             error("Failed to create running execution record: #{e.message}")
             nil
@@ -120,14 +127,10 @@ module RubyLLM
             end
 
             update_data = build_completion_data(context, status)
+            execution.update!(update_data)
 
-            if async_logging?
-              # For async updates, use a job (if update support exists)
-              # For now, update synchronously to ensure dashboard shows correct status
-              execution.update!(update_data)
-            else
-              execution.update!(update_data)
-            end
+            # Save detail data (prompts, responses, tool calls, etc.)
+            save_execution_details(execution, context, status)
           rescue StandardError => e
             error("Failed to complete execution record: #{e.message}")
             raise # Re-raise for ensure block to handle via mark_execution_failed!
@@ -150,11 +153,22 @@ module RubyLLM
             update_data = {
               status: "error",
               completed_at: Time.current,
-              error_class: error&.class&.name || "UnknownError",
-              error_message: error_message
+              error_class: error&.class&.name || "UnknownError"
             }
 
             execution.class.where(id: execution.id, status: "running").update_all(update_data)
+
+            # Store error_message in detail table (best-effort)
+            begin
+              detail_attrs = { error_message: error_message }
+              if execution.detail
+                execution.detail.update_columns(detail_attrs)
+              else
+                RubyLLM::Agents::ExecutionDetail.create!(detail_attrs.merge(execution_id: execution.id))
+              end
+            rescue StandardError
+              # Non-critical
+            end
           rescue StandardError => e
             error("CRITICAL: Failed emergency status update for execution #{execution&.id}: #{e.message}")
           end
@@ -174,7 +188,6 @@ module RubyLLM
           def build_running_execution_data(context)
             data = {
               agent_type: context.agent_class&.name,
-              agent_version: config(:version, "1.0"),
               model_id: context.model,
               status: "running",
               started_at: context.started_at,
@@ -189,9 +202,6 @@ module RubyLLM
               data[:tenant_id] = context.tenant_id
             end
 
-            # Add sanitized parameters
-            data[:parameters] = sanitize_parameters(context)
-
             data
           end
 
@@ -212,38 +222,63 @@ module RubyLLM
               attempts_count: context.attempts_made
             }
 
-            # Add cache key for cache hit executions
+            # Store niche cache key in metadata
+            merged_metadata = context.metadata.dup rescue {}
             if context.cached? && context[:cache_key]
-              data[:response_cache_key] = context[:cache_key]
+              merged_metadata["response_cache_key"] = context[:cache_key]
             end
+            data[:metadata] = merged_metadata if merged_metadata.any?
 
-            # Add error details if present
+            # Error class on execution (error_message goes to detail)
             if context.error
               data[:error_class] = context.error.class.name
-              data[:error_message] = truncate_error_message(context.error.message)
             end
 
-            # Add custom metadata
-            data[:metadata] = context.metadata if context.metadata.any?
-
-            # Add enhanced tool calls if present
+            # Tool calls count on execution
             if context[:tool_calls].present?
-              data[:tool_calls] = context[:tool_calls]
               data[:tool_calls_count] = context[:tool_calls].size
             end
 
-            # Add reliability attempts if present
+            # Attempts count on execution
             if context[:reliability_attempts].present?
-              data[:attempts] = context[:reliability_attempts]
               data[:attempts_count] = context[:reliability_attempts].size
             end
 
-            # Add response if persist_responses is enabled
+            data
+          end
+
+          # Saves detail data to the execution_details table after completion
+          def save_execution_details(execution, context, status)
+            return unless execution
+
+            detail_data = {}
+
+            if context.error
+              detail_data[:error_message] = truncate_error_message(context.error.message)
+            end
+
+            if context[:tool_calls].present?
+              detail_data[:tool_calls] = context[:tool_calls]
+            end
+
+            if context[:reliability_attempts].present?
+              detail_data[:attempts] = context[:reliability_attempts]
+            end
+
             if global_config.persist_responses && context.output.respond_to?(:content)
-              data[:response] = serialize_response(context)
+              detail_data[:response] = serialize_response(context)
             end
 
-            data
+            has_data = detail_data.values.any? { |v| v.present? && v != {} && v != [] }
+            return unless has_data
+
+            if execution.detail
+              execution.detail.update!(detail_data)
+            else
+              execution.create_detail!(detail_data)
+            end
+          rescue StandardError => e
+            error("Failed to save execution details: #{e.message}")
           end
 
           # Persists execution data to database (legacy fallback)
@@ -272,9 +307,13 @@ module RubyLLM
           # @param status [String] "success" or "error"
           # @return [Hash] Execution data
           def build_execution_data(context, status)
+            merged_metadata = context.metadata.dup rescue {}
+            if context.cached? && context[:cache_key]
+              merged_metadata["response_cache_key"] = context[:cache_key]
+            end
+
             data = {
               agent_type: context.agent_class&.name,
-              agent_version: config(:version, "1.0"),
               model_id: context.model,
               status: determine_status(context, status),
               duration_ms: context.duration_ms,
@@ -284,7 +323,8 @@ module RubyLLM
               input_tokens: context.input_tokens || 0,
               output_tokens: context.output_tokens || 0,
               total_cost: context.total_cost || 0,
-              attempts_count: context.attempts_made
+              attempts_count: context.attempts_made,
+              metadata: merged_metadata
             }
 
             # Add tenant_id only if multi-tenancy is enabled and tenant is set
@@ -292,39 +332,30 @@ module RubyLLM
               data[:tenant_id] = context.tenant_id
             end
 
-            # Add cache key for cache hit executions
-            if context.cached? && context[:cache_key]
-              data[:response_cache_key] = context[:cache_key]
-            end
-
-            # Add error details if present
+            # Error class on execution
             if context.error
               data[:error_class] = context.error.class.name
-              data[:error_message] = truncate_error_message(context.error.message)
             end
 
-            # Add custom metadata
-            data[:metadata] = context.metadata if context.metadata.any?
-
-            # Add sanitized parameters
-            data[:parameters] = sanitize_parameters(context)
-
-            # Add enhanced tool calls if present
+            # Tool calls count on execution
             if context[:tool_calls].present?
-              data[:tool_calls] = context[:tool_calls]
               data[:tool_calls_count] = context[:tool_calls].size
             end
 
-            # Add reliability attempts if present
+            # Attempts count on execution
             if context[:reliability_attempts].present?
-              data[:attempts] = context[:reliability_attempts]
               data[:attempts_count] = context[:reliability_attempts].size
             end
 
-            # Add response if persist_responses is enabled
+            # Store detail data for separate creation
+            detail_data = { parameters: sanitize_parameters(context) }
+            detail_data[:error_message] = truncate_error_message(context.error.message) if context.error
+            detail_data[:tool_calls] = context[:tool_calls] if context[:tool_calls].present?
+            detail_data[:attempts] = context[:reliability_attempts] if context[:reliability_attempts].present?
             if global_config.persist_responses && context.output.respond_to?(:content)
-              data[:response] = serialize_response(context)
+              detail_data[:response] = serialize_response(context)
             end
+            data[:_detail_data] = detail_data
 
             data
           end
@@ -401,8 +432,7 @@ module RubyLLM
             response_data[:input_tokens] = context.input_tokens if context.input_tokens
             response_data[:output_tokens] = context.output_tokens if context.output_tokens
 
-            # Apply redaction for sensitive data
-            Redactor.redact(response_data)
+            response_data
           rescue StandardError => e
             error("Failed to serialize response: #{e.message}")
             nil
@@ -419,7 +449,12 @@ module RubyLLM
           #
           # @param data [Hash] Execution data
           def create_execution_record(data)
-            Execution.create!(data)
+            detail_data = data.delete(:_detail_data)
+            execution = Execution.create!(data)
+            if detail_data && detail_data.values.any? { |v| v.present? && v != {} && v != [] }
+              execution.create_detail!(detail_data)
+            end
+            execution
           end
 
           # Returns whether tracking is enabled for this agent type
@@ -432,8 +467,6 @@ module RubyLLM
             case context.agent_type
             when :embedding
               cfg.track_embeddings
-            when :moderation
-              cfg.track_moderation
             when :image
               cfg.track_image_generation
             when :audio

data/lib/ruby_llm/agents/pipeline/middleware/tenant.rb

@@ -8,18 +8,16 @@ module RubyLLM
         #
         # This middleware extracts tenant information from the context options,
         # sets the tenant_id, tenant_object, and tenant_config on the context,
-        # and applies the resolved API configuration to RubyLLM.
+        # and applies any tenant-specific API keys to RubyLLM.
         #
         # Supports three formats:
         # - Object with llm_tenant_id method (recommended for ActiveRecord models)
         # - Hash with :id key (simple/legacy format)
         # - nil (no tenant - single-tenant mode)
         #
-        # API key resolution priority:
-        # 1. Tenant object's llm_api_keys method (if present)
-        # 2. Tenant-specific database config (ApiConfiguration)
-        # 3. Global database config
-        # 4. RubyLLM.configuration (set via initializer or environment)
+        # API keys are configured via:
+        # - RubyLLM.configuration (set via initializer or environment variables)
+        # - Tenant object's llm_api_keys method (for per-tenant overrides)
         #
         # @example With ActiveRecord model
         #   # Model uses llm_tenant DSL
@@ -88,16 +86,10 @@ module RubyLLM
 
           # Applies API configuration to RubyLLM based on resolved tenant
           #
-          # This method resolves API keys from multiple sources and applies
-          # them to RubyLLM.config before the agent executes.
-          #
           # @param context [Context] The execution context
           def apply_api_configuration!(context)
-            # First, try to apply keys from tenant object's llm_api_keys method
+            # Apply keys from tenant object's llm_api_keys method if present
             apply_tenant_object_api_keys!(context)
-
-            # Then, apply database configuration (tenant > global > ruby_llm_config)
-            apply_database_api_configuration!(context)
           end
 
           # Applies API keys from tenant object's llm_api_keys method
@@ -116,22 +108,6 @@ module RubyLLM
             warn_api_key_error("tenant object", e)
           end
 
-          # Applies API configuration from the database
-          #
-          # @param context [Context] The execution context
-          def apply_database_api_configuration!(context)
-            return unless api_configuration_available?
-
-            resolved = ApiConfiguration.resolve(tenant_id: context.tenant_id)
-            resolved.apply_to_ruby_llm!
-
-            # Store resolved config on context for observability
-            context[:resolved_api_config] = resolved
-          rescue StandardError => e
-            # Log but don't fail if DB lookup fails
-            warn_api_key_error("database", e)
-          end
-
           # Applies a hash of API keys to RubyLLM configuration
           #
           # @param api_keys [Hash] Hash of provider => key mappings
@@ -154,18 +130,6 @@ module RubyLLM
             "#{provider}_api_key="
           end
 
-          # Checks if ApiConfiguration model is available
-          #
-          # @return [Boolean]
-          def api_configuration_available?
-            return false unless defined?(RubyLLM::Agents::ApiConfiguration)
-
-            # Check if table exists
-            ApiConfiguration.table_exists?
-          rescue StandardError
-            false
-          end
-
           # Logs a warning about API key resolution failure
           #
           # @param source [String] Source that failed

data/lib/ruby_llm/agents/rails/engine.rb

@@ -34,9 +34,7 @@ module RubyLLM
       config.to_prepare do
         require_relative "../infrastructure/execution_logger_job"
         require_relative "../core/instrumentation"
-        require_relative "../core/resolved_config"
         require_relative "../core/base"
-        require_relative "../workflow/orchestrator"
 
         # Resolve the parent controller class from configuration
         # Default is ActionController::Base, but can be set to inherit from app controllers
@@ -161,6 +159,12 @@ module RubyLLM
         end)
       end
 
+      # Load rake tasks from lib/tasks
+      rake_tasks do
+        tasks_path = File.expand_path("../../../tasks", __dir__)
+        Dir[File.join(tasks_path, "**", "*.rake")].each { |f| load f }
+      end
+
       # Configures default generators for the engine
       # Sets up RSpec and FactoryBot for generated specs
       # @api private
@@ -179,7 +183,6 @@ module RubyLLM
       # - app/agents/ (top-level, no namespace)
       # - app/agents/embedders/ -> Embedders namespace
       # - app/agents/images/ -> Images namespace
-      # - app/workflows/ (top-level, no namespace)
       #
       # @api private
       initializer "ruby_llm_agents.autoload_agents", before: :set_autoload_paths do |app|
@@ -219,9 +222,6 @@ module RubyLLM
       def self.namespace_for_path(path, config)
         parts = path.split("/")
 
-        # app/workflows -> no namespace (top-level workflows)
-        return nil if parts == ["app", "workflows"]
-
         # Need at least app/{root_directory}
         return nil unless parts.length >= 2 && parts[0] == "app"
         return nil unless parts[1] == config.root_directory

data/lib/ruby_llm/agents/results/base.rb

@@ -102,15 +102,6 @@ module RubyLLM
       #   @return [Integer, nil] Number of tokens used for thinking
       attr_reader :thinking_text, :thinking_signature, :thinking_tokens
 
-      # @!group Moderation
-      # @!attribute [r] status
-      #   @return [Symbol, nil] Result status (:success, :input_moderation_blocked, :output_moderation_blocked)
-      # @!attribute [r] moderation_result
-      #   @return [Object, nil] The raw moderation result from RubyLLM
-      # @!attribute [r] moderation_phase
-      #   @return [Symbol, nil] The phase where moderation blocked (:input or :output)
-      attr_reader :status, :moderation_result, :moderation_phase
-
       # Creates a new Result instance
       #
       # @param content [Hash, String] The processed response content
@@ -160,12 +151,6 @@ module RubyLLM
         @thinking_text = options[:thinking_text]
         @thinking_signature = options[:thinking_signature]
         @thinking_tokens = options[:thinking_tokens]
-
-        # Moderation
-        @status = options[:status] || :success
-        @moderation_flagged = options[:moderation_flagged] || false
-        @moderation_result = options[:moderation_result]
-        @moderation_phase = options[:moderation_phase]
       end
 
       # Returns total tokens (input + output)
@@ -224,34 +209,6 @@ module RubyLLM
         thinking_text.present?
       end
 
-      # Returns whether content was flagged by moderation
-      #
-      # @return [Boolean] true if moderation flagged the content
-      def moderation_flagged?
-        @moderation_flagged == true
-      end
-
-      # Returns whether content passed moderation
-      #
-      # @return [Boolean] true if content was not flagged
-      def moderation_passed?
-        !moderation_flagged?
-      end
-
-      # Returns the categories flagged by moderation
-      #
-      # @return [Array<String, Symbol>] Flagged category names
-      def moderation_categories
-        @moderation_result&.flagged_categories || []
-      end
-
-      # Returns the moderation category scores
-      #
-      # @return [Hash{String, Symbol => Float}] Category to score mapping
-      def moderation_scores
-        @moderation_result&.category_scores || {}
-      end
-
       # Converts the result to a hash
       #
       # @return [Hash] All result data as a hash
@@ -283,12 +240,7 @@ module RubyLLM
           tool_calls_count: tool_calls_count,
           thinking_text: thinking_text,
           thinking_signature: thinking_signature,
-          thinking_tokens: thinking_tokens,
-          status: status,
-          moderation_flagged: moderation_flagged?,
-          moderation_phase: moderation_phase,
-          moderation_categories: moderation_categories,
-          moderation_scores: moderation_scores
+          thinking_tokens: thinking_tokens
         }
       end
 

data/lib/ruby_llm/agents/text/embedder.rb

@@ -257,7 +257,6 @@ module RubyLLM
           "ruby_llm_agents",
           "embedding",
           self.class.name,
-          self.class.version,
           resolved_model,
           resolved_dimensions,
           Digest::SHA256.hexdigest(input_texts.map { |t| preprocess(t) }.join("\n"))

data/lib/ruby_llm/agents.rb

@@ -2,13 +2,13 @@
 
 require "csv"
 require "ruby_llm"
+require "ruby_llm/schema"
 
 # Core
 require_relative "agents/core/version"
 require_relative "agents/core/configuration"
 require_relative "agents/core/deprecations"
 require_relative "agents/core/errors"
-require_relative "agents/core/resolved_config"
 require_relative "agents/core/llm_tenant"
 
 # Infrastructure - Reliability
@@ -29,7 +29,6 @@ require_relative "agents/dsl"
 require_relative "agents/base_agent"
 
 # Infrastructure - Budget & Utilities
-require_relative "agents/infrastructure/redactor"
 require_relative "agents/infrastructure/circuit_breaker"
 require_relative "agents/infrastructure/budget_tracker"
 require_relative "agents/infrastructure/alert_manager"
@@ -43,7 +42,6 @@ require_relative "agents/infrastructure/budget/spend_recorder"
 # Results
 require_relative "agents/results/base"
 require_relative "agents/results/embedding_result"
-require_relative "agents/results/moderation_result"
 require_relative "agents/results/transcription_result"
 require_relative "agents/results/speech_result"
 require_relative "agents/results/image_generation_result"
@@ -61,7 +59,6 @@ require_relative "agents/image/concerns/image_operation_execution"
 
 # Text agents
 require_relative "agents/text/embedder"
-require_relative "agents/text/moderator"
 
 # Audio agents
 require_relative "agents/audio/transcriber"
@@ -77,11 +74,6 @@ require_relative "agents/image/analyzer"
 require_relative "agents/image/background_remover"
 require_relative "agents/image/pipeline"
 
-# Workflow
-require_relative "agents/workflow/async"
-require_relative "agents/workflow/orchestrator"
-require_relative "agents/workflow/async_executor"
-
 # Rails integration
 if defined?(Rails)
   require_relative "agents/core/inflections"

data/lib/tasks/ruby_llm_agents.rake

@@ -0,0 +1,34 @@
+# frozen_string_literal: true
+
+namespace :ruby_llm_agents do
+  namespace :tenants do
+    desc "Refresh all tenant counters from executions table"
+    task refresh: :environment do
+      count = 0
+      RubyLLM::Agents::Tenant.find_each do |tenant|
+        tenant.refresh_counters!
+        count += 1
+      end
+      puts "Refreshed #{count} tenants"
+    end
+
+    desc "Refresh active tenant counters from executions table"
+    task refresh_active: :environment do
+      count = 0
+      RubyLLM::Agents::Tenant.active.find_each do |tenant|
+        tenant.refresh_counters!
+        count += 1
+      end
+      puts "Refreshed #{count} active tenants"
+    end
+
+    desc "Refresh a single tenant's counters"
+    task :refresh_one, [:tenant_id] => :environment do |_, args|
+      abort "Usage: rake ruby_llm_agents:tenants:refresh_one[tenant_id]" unless args[:tenant_id]
+
+      tenant = RubyLLM::Agents::Tenant.find_by!(tenant_id: args[:tenant_id])
+      tenant.refresh_counters!
+      puts "Refreshed tenant: #{tenant.tenant_id}"
+    end
+  end
+end