ruby_llm-agents 3.3.0 → 3.5.0

data/lib/ruby_llm/agents/audio/speech_pricing.rb

@@ -8,10 +8,11 @@ module RubyLLM
     module Audio
       # Dynamic pricing resolution for text-to-speech models.
       #
-      # Uses the same three-tier strategy as ImageGenerator::Pricing:
+      # Uses a four-tier pricing cascade:
       # 1. LiteLLM JSON (primary) - future-proof, auto-updating
       # 2. Configurable pricing table - user overrides via config.tts_model_pricing
-      # 3. Hardcoded fallbacks - per-model defaults
+      # 3. ElevenLabs API - dynamic multiplier × base rate from /v1/models
+      # 4. Hardcoded fallbacks - per-model defaults
       #
       # All prices are per 1,000 characters.
       #
@@ -50,14 +51,22 @@ module RubyLLM
         # @param model_id [String] Model identifier
         # @return [Float] Cost per 1K characters in USD
         def cost_per_1k_characters(provider, model_id)
+          # Tier 1: LiteLLM
           if (litellm_price = from_litellm(model_id))
             return litellm_price
           end
 
+          # Tier 2: User config overrides
           if (config_price = from_config(model_id))
             return config_price
           end
 
+          # Tier 3: ElevenLabs API multiplier × base rate
+          if provider == :elevenlabs && (api_price = from_elevenlabs_api(model_id))
+            return api_price
+          end
+
+          # Tier 4: Hardcoded fallbacks
           fallback_price(provider, model_id)
         end
 
@@ -73,6 +82,7 @@ module RubyLLM
           {
             litellm: litellm_tts_models,
             configured: config.tts_model_pricing || {},
+            elevenlabs_api: elevenlabs_api_pricing,
             fallbacks: fallback_pricing_table
           }
         end
@@ -190,6 +200,19 @@ module RubyLLM
           end
         end
 
+        def elevenlabs_api_pricing
+          return {} unless defined?(ElevenLabs::ModelRegistry)
+
+          base = config.elevenlabs_base_cost_per_1k || 0.30
+          ElevenLabs::ModelRegistry.models.each_with_object({}) do |model, hash|
+            multiplier = model.dig("model_rates", "character_cost_multiplier") || 1.0
+            hash[model["model_id"]] = (base * multiplier).round(6)
+          end
+        rescue => e
+          warn "[RubyLLM::Agents] Failed to get ElevenLabs API pricing: #{e.message}"
+          {}
+        end
+
         # ============================================================
         # Tier 2: User configuration
         # ============================================================
@@ -207,7 +230,25 @@ module RubyLLM
         end
 
         # ============================================================
-        # Tier 3: Hardcoded fallbacks
+        # Tier 3: ElevenLabs API (dynamic multiplier × base rate)
+        # ============================================================
+
+        def from_elevenlabs_api(model_id)
+          return nil unless defined?(ElevenLabs::ModelRegistry)
+
+          model = ElevenLabs::ModelRegistry.find(model_id)
+          return nil unless model
+
+          multiplier = model.dig("model_rates", "character_cost_multiplier") || 1.0
+          base = config.elevenlabs_base_cost_per_1k || 0.30
+          (base * multiplier).round(6)
+        rescue => e
+          warn "[RubyLLM::Agents] Failed to get ElevenLabs API pricing: #{e.message}"
+          nil
+        end
+
+        # ============================================================
+        # Tier 4: Hardcoded fallbacks
         # ============================================================
 
         def fallback_price(provider, model_id)

data/lib/ruby_llm/agents/audio/transcriber.rb

@@ -2,6 +2,7 @@
 
 require "digest"
 require_relative "../results/transcription_result"
+require_relative "transcription_pricing"
 
 module RubyLLM
   module Agents
@@ -318,6 +319,12 @@ module RubyLLM
         context.output_tokens = 0
         context.total_cost = calculate_cost(raw_result)
 
+        # Store pricing warning if cost calculation returned nil
+        if @pricing_warning
+          context[:pricing_warning] = @pricing_warning
+          Rails.logger.warn(@pricing_warning) if defined?(Rails) && Rails.respond_to?(:logger) && Rails.logger
+        end
+
         # Store transcription-specific metadata for execution tracking
         context[:language] = resolved_language if resolved_language
         context[:detected_language] = raw_result[:language] if raw_result[:language]
@@ -615,30 +622,34 @@ module RubyLLM
       # Calculates cost for transcription
       #
       # @param raw_result [Hash] Raw transcription result
-      # @return [Float] Cost in USD
+      # @return [Float] Cost in USD (0 if no pricing found)
       def calculate_cost(raw_result)
-        # Get duration in minutes
-        duration_minutes = raw_result[:duration] ? raw_result[:duration] / 60.0 : 0
+        @pricing_warning = nil
 
-        # Check if response has cost info
+        # Check if response has cost info from the API
         if raw_result[:raw_response].respond_to?(:cost) && raw_result[:raw_response].cost
           return raw_result[:raw_response].cost
         end
 
-        # Estimate based on model and duration
+        # Delegate to TranscriptionPricing (2-tier: LiteLLM + user config)
         model = raw_result[:model].to_s
-        price_per_minute = case model
-        when /whisper-1/
-          0.006
-        when /gpt-4o-transcribe/
-          0.01
-        when /gpt-4o-mini-transcribe/
-          0.005
-        else
-          0.006 # Default to whisper pricing
+        duration = raw_result[:duration] || 0
+
+        cost = Audio::TranscriptionPricing.calculate_cost(
+          model_id: model,
+          duration_seconds: duration
+        )
+
+        if cost.nil?
+          @pricing_warning = "[RubyLLM::Agents] No pricing found for transcription model '#{model}'. " \
+            "Cost recorded as $0. Add pricing to your config:\n" \
+            "  RubyLLM::Agents.configure do |c|\n" \
+            "    c.transcription_model_pricing = { \"#{model}\" => 0.006 }  # price per minute\n" \
+            "  end"
+          return 0
         end
 
-        duration_minutes * price_per_minute
+        cost
       end
 
       # Resolves the model to use

data/lib/ruby_llm/agents/audio/transcription_pricing.rb

@@ -0,0 +1,226 @@
+# frozen_string_literal: true
+
+require_relative "../pricing/data_store"
+require_relative "../pricing/ruby_llm_adapter"
+require_relative "../pricing/litellm_adapter"
+require_relative "../pricing/portkey_adapter"
+require_relative "../pricing/openrouter_adapter"
+require_relative "../pricing/helicone_adapter"
+require_relative "../pricing/llmpricing_adapter"
+
+module RubyLLM
+  module Agents
+    module Audio
+      # Dynamic pricing resolution for audio transcription models.
+      #
+      # Cascades through multiple pricing sources to maximize coverage:
+      # 1. User config (instant, always wins)
+      # 2. RubyLLM gem (local, no HTTP, already a dependency)
+      # 3. LiteLLM (bulk, most comprehensive for transcription)
+      # 4. Portkey AI (per-model, good transcription coverage)
+      # 5. OpenRouter (bulk, audio-capable chat models only)
+      # 6. Helicone (text LLM only — pass-through, future-proof)
+      # 7. LLM Pricing AI (text LLM only — pass-through, future-proof)
+      #
+      # When no pricing is found, methods return nil to signal the caller
+      # should warn the user with actionable configuration instructions.
+      #
+      # All prices are per minute of audio.
+      #
+      # @example Get cost for a transcription
+      #   TranscriptionPricing.calculate_cost(model_id: "whisper-1", duration_seconds: 120)
+      #   # => 0.012 (or nil if no pricing found)
+      #
+      # @example User-configured pricing
+      #   RubyLLM::Agents.configure do |c|
+      #     c.transcription_model_pricing = { "whisper-1" => 0.006 }
+      #   end
+      #
+      module TranscriptionPricing
+        extend self
+
+        LITELLM_PRICING_URL = Pricing::DataStore::LITELLM_URL
+
+        SOURCES = [:config, :ruby_llm, :litellm, :portkey, :openrouter, :helicone, :llmpricing].freeze
+
+        # Calculate total cost for a transcription operation
+        #
+        # @param model_id [String] The model identifier
+        # @param duration_seconds [Numeric] Duration of audio in seconds
+        # @return [Float, nil] Total cost in USD, or nil if no pricing found
+        def calculate_cost(model_id:, duration_seconds:)
+          price = cost_per_minute(model_id)
+          return nil unless price
+
+          duration_minutes = duration_seconds / 60.0
+          (duration_minutes * price).round(6)
+        end
+
+        # Get cost per minute for a transcription model
+        #
+        # @param model_id [String] Model identifier
+        # @return [Float, nil] Cost per minute in USD, or nil if not found
+        def cost_per_minute(model_id)
+          SOURCES.each do |source|
+            price = send(:"from_#{source}", model_id)
+            return price if price
+          end
+          nil
+        end
+
+        # Check whether pricing is available for a model
+        #
+        # @param model_id [String] Model identifier
+        # @return [Boolean] true if pricing is available
+        def pricing_found?(model_id)
+          !cost_per_minute(model_id).nil?
+        end
+
+        # Force refresh of cached pricing data
+        def refresh!
+          Pricing::DataStore.refresh!
+        end
+
+        # Expose all known pricing for debugging/dashboard
+        #
+        # @return [Hash] Pricing from all tiers
+        def all_pricing
+          {
+            ruby_llm: {},  # local gem, per-model lookup
+            litellm: litellm_transcription_models,
+            portkey: {},  # per-model, populated on demand
+            openrouter: {},  # no dedicated transcription models
+            helicone: {},  # no transcription models
+            configured: config.transcription_model_pricing || {}
+          }
+        end
+
+        private
+
+        # ============================================================
+        # Tier 1: User configuration (highest priority)
+        # ============================================================
+
+        def from_config(model_id)
+          table = config.transcription_model_pricing
+          return nil unless table.is_a?(Hash) && !table.empty?
+
+          normalized = normalize_model_id(model_id)
+
+          price = table[model_id] || table[normalized] ||
+            table[model_id.to_sym] || table[normalized.to_sym]
+
+          price if price.is_a?(Numeric)
+        end
+
+        # ============================================================
+        # Tier 2: RubyLLM gem (local, no HTTP)
+        # ============================================================
+
+        def from_ruby_llm(model_id)
+          data = Pricing::RubyLLMAdapter.find_model(model_id)
+          return nil unless data
+
+          extract_per_minute(data)
+        end
+
+        # ============================================================
+        # Tier 3: LiteLLM
+        # ============================================================
+
+        def from_litellm(model_id)
+          data = Pricing::LiteLLMAdapter.find_model(model_id)
+          return nil unless data
+
+          extract_per_minute(data)
+        end
+
+        # ============================================================
+        # Tier 4: Portkey AI
+        # ============================================================
+
+        def from_portkey(model_id)
+          data = Pricing::PortkeyAdapter.find_model(model_id)
+          return nil unless data
+
+          extract_per_minute(data)
+        end
+
+        # ============================================================
+        # Tier 5: OpenRouter (audio-capable chat models only)
+        # ============================================================
+
+        def from_openrouter(model_id)
+          data = Pricing::OpenRouterAdapter.find_model(model_id)
+          return nil unless data
+
+          extract_per_minute(data)
+        end
+
+        # ============================================================
+        # Tier 6: Helicone (text LLM only — future-proof)
+        # ============================================================
+
+        def from_helicone(model_id)
+          data = Pricing::HeliconeAdapter.find_model(model_id)
+          return nil unless data
+
+          extract_per_minute(data)
+        end
+
+        # ============================================================
+        # Tier 7: LLM Pricing AI (text LLM only — future-proof)
+        # ============================================================
+
+        def from_llmpricing(model_id)
+          data = Pricing::LLMPricingAdapter.find_model(model_id)
+          return nil unless data
+
+          extract_per_minute(data)
+        end
+
+        # ============================================================
+        # Price extraction
+        # ============================================================
+
+        def extract_per_minute(data)
+          # Per-second pricing (most common for transcription: whisper-1, etc.)
+          if data[:input_cost_per_second]
+            return (data[:input_cost_per_second] * 60).round(6)
+          end
+
+          # Per-audio-token pricing (GPT-4o-transcribe models)
+          # ~25 audio tokens/second = 1500 tokens/minute
+          if data[:input_cost_per_audio_token]
+            return (data[:input_cost_per_audio_token] * 1500).round(6)
+          end
+
+          nil
+        end
+
+        def litellm_transcription_models
+          data = Pricing::DataStore.litellm_data
+          return {} unless data.is_a?(Hash)
+
+          data.select do |key, value|
+            value.is_a?(Hash) && (
+              value["mode"] == "audio_transcription" ||
+              value["input_cost_per_second"] ||
+              key.to_s.match?(/whisper|transcri/i)
+            )
+          end
+        end
+
+        def normalize_model_id(model_id)
+          model_id.to_s.downcase
+            .gsub(/[^a-z0-9._-]/, "-").squeeze("-")
+            .gsub(/^-|-$/, "")
+        end
+
+        def config
+          RubyLLM::Agents.configuration
+        end
+      end
+    end
+  end
+end

data/lib/ruby_llm/agents/core/configuration.rb

@@ -453,7 +453,20 @@ module RubyLLM
         :root_namespace,
         :tool_result_max_length,
         :redaction,
-        :persist_audio_data
+        :persist_audio_data,
+        :elevenlabs_base_cost_per_1k,
+        :elevenlabs_models_cache_ttl,
+        :transcription_model_pricing,
+        :default_transcription_cost,
+        :pricing_cache_ttl,
+        :portkey_pricing_enabled,
+        :portkey_pricing_url,
+        :openrouter_pricing_enabled,
+        :openrouter_pricing_url,
+        :helicone_pricing_enabled,
+        :helicone_pricing_url,
+        :llmpricing_enabled,
+        :llmpricing_url
 
       # Attributes with validation (readers only, custom setters below)
       attr_reader :default_temperature,
@@ -672,6 +685,19 @@ module RubyLLM
         # Transcription defaults
         @default_transcription_model = "whisper-1"
         @track_transcriptions = true
+        @transcription_model_pricing = {}
+        @default_transcription_cost = nil  # nil = no default, will trigger warning
+
+        # Multi-source pricing defaults
+        @pricing_cache_ttl = nil  # nil = use DataStore default (24h)
+        @portkey_pricing_enabled = true
+        @portkey_pricing_url = nil  # nil = use default
+        @openrouter_pricing_enabled = true
+        @openrouter_pricing_url = nil
+        @helicone_pricing_enabled = true
+        @helicone_pricing_url = nil
+        @llmpricing_enabled = true
+        @llmpricing_url = nil
 
         # TTS/Speech defaults
         @default_tts_provider = :openai
@@ -738,6 +764,11 @@ module RubyLLM
 
         # Audio data persistence (disabled by default — base64 audio can be large)
         @persist_audio_data = false
+
+        # ElevenLabs dynamic pricing: base cost per 1K characters (Pro plan overage rate)
+        @elevenlabs_base_cost_per_1k = 0.30
+        # ElevenLabs models cache TTL in seconds (6 hours)
+        @elevenlabs_models_cache_ttl = 21_600
       end
 
       # Returns the configured cache store, falling back to Rails.cache

data/lib/ruby_llm/agents/core/version.rb

@@ -4,6 +4,6 @@ module RubyLLM
   module Agents
     # Current version of the RubyLLM::Agents gem
     # @return [String] Semantic version string
-    VERSION = "3.3.0"
+    VERSION = "3.5.0"
   end
 end