kreuzberg 4.0.0.rc1 → 4.0.0.rc2

data/lib/kreuzberg/config.rb

@@ -1,684 +1,691 @@
-# frozen_string_literal: true
-
-module Kreuzberg
-  module Config
-    # OCR configuration
-    #
-    # @example
-    #   ocr = OCR.new(backend: "tesseract", language: "eng")
-    #
-    class OCR
-      attr_reader :backend, :language, :tesseract_config
-
-      def initialize(
-        backend: 'tesseract',
-        language: 'eng',
-        tesseract_config: nil
-      )
-        @backend = backend.to_s
-        @language = language.to_s
-        @tesseract_config = normalize_tesseract_config(tesseract_config)
-      end
-
-      def to_h
-        {
-          backend: @backend,
-          language: @language,
-          tesseract_config: @tesseract_config&.to_h
-        }.compact
-      end
-
-      private
-
-      def normalize_tesseract_config(value)
-        return nil if value.nil?
-        return value if value.is_a?(Tesseract)
-        return Tesseract.new(**value.transform_keys(&:to_sym)) if value.is_a?(Hash)
-
-        raise ArgumentError, "Expected #{Tesseract}, Hash, or nil, got #{value.class}"
-      end
-    end
-
-    class Tesseract
-      attr_reader :options
-
-      def initialize(**options)
-        @options = options.transform_keys(&:to_sym)
-        normalize_nested_preprocessing!
-      end
-
-      def to_h
-        @options.dup
-      end
-
-      private
-
-      def normalize_nested_preprocessing!
-        preprocessing = @options[:preprocessing]
-        return if preprocessing.nil?
-        return if preprocessing.is_a?(ImagePreprocessing)
-        return @options[:preprocessing] = ImagePreprocessing.new(**preprocessing.transform_keys(&:to_sym)) if
-          preprocessing.is_a?(Hash)
-
-        raise ArgumentError, "preprocessing must be #{ImagePreprocessing} or Hash"
-      end
-    end
-
-    # Chunking configuration
-    #
-    # @example
-    #   chunking = Chunking.new(max_chars: 1000, max_overlap: 200)
-    #
-    class Chunking
-      attr_reader :max_chars, :max_overlap, :preset, :embedding, :enabled
-
-      def initialize(
-        max_chars: nil,
-        max_overlap: nil,
-        preset: nil,
-        embedding: nil,
-        chunk_size: nil,
-        chunk_overlap: nil,
-        enabled: true
-      )
-        resolved_size = chunk_size || max_chars || 1000
-        resolved_overlap = chunk_overlap || max_overlap || 200
-
-        @max_chars = resolved_size.to_i
-        @max_overlap = resolved_overlap.to_i
-        @preset = preset&.to_s
-        @embedding = normalize_embedding(embedding)
-        @enabled = boolean_or_nil(enabled)
-      end
-
-      def to_h
-        config = {
-          max_chars: @max_chars,
-          max_overlap: @max_overlap,
-          preset: @preset,
-          embedding: @embedding&.to_h
-        }.compact
-        # @type var config: Hash[Symbol, untyped]
-        config[:enabled] = @enabled unless @enabled.nil?
-        config
-      end
-
-      private
-
-      def normalize_embedding(value)
-        return nil if value.nil?
-        return value if value.is_a?(Embedding)
-        return Embedding.new(**value.transform_keys(&:to_sym)) if value.is_a?(Hash)
-
-        raise ArgumentError, "Expected #{Embedding}, Hash, or nil, got #{value.class}"
-      end
-
-      def boolean_or_nil(value)
-        return nil if value.nil?
-
-        value ? true : false
-      end
-    end
-
-    class Embedding
-      attr_reader :model, :normalize, :batch_size, :show_download_progress, :cache_dir
-
-      def initialize(
-        model: { type: :preset, name: 'balanced' },
-        normalize: true,
-        batch_size: 32,
-        show_download_progress: false,
-        cache_dir: nil
-      )
-        @model = normalize_model(model)
-        @normalize = boolean_or_nil(normalize)
-        @batch_size = batch_size&.to_i
-        @show_download_progress = boolean_or_nil(show_download_progress)
-        @cache_dir = cache_dir&.to_s
-      end
-
-      def to_h
-        {
-          model: @model,
-          normalize: @normalize,
-          batch_size: @batch_size,
-          show_download_progress: @show_download_progress,
-          cache_dir: @cache_dir
-        }.compact
-      end
-
-      private
-
-      def normalize_model(model)
-        normalized = if model.respond_to?(:to_h)
-                       model.to_h
-                     else
-                       model
-                     end
-        raise ArgumentError, 'model must be a Hash describing the embedding model' unless normalized.is_a?(Hash)
-
-        normalized.transform_keys(&:to_sym)
-      end
-
-      def boolean_or_nil(value)
-        return nil if value.nil?
-
-        value ? true : false
-      end
-    end
-
-    # Language detection configuration
-    #
-    # @example
-    #   lang = LanguageDetection.new(enabled: true, min_confidence: 0.8)
-    #
-    class LanguageDetection
-      attr_reader :enabled, :min_confidence, :detect_multiple
-
-      def initialize(enabled: false, min_confidence: 0.5, detect_multiple: false)
-        @enabled = enabled ? true : false
-        @min_confidence = min_confidence.to_f
-        @detect_multiple = detect_multiple ? true : false
-      end
-
-      def to_h
-        {
-          enabled: @enabled,
-          min_confidence: @min_confidence,
-          detect_multiple: @detect_multiple
-        }
-      end
-    end
-
-    # PDF-specific options
-    #
-    # @example
-    #   pdf = PDF.new(extract_images: true, passwords: ["secret", "backup"])
-    #
-    class PDF
-      attr_reader :extract_images, :passwords, :extract_metadata
-
-      def initialize(
-        extract_images: false,
-        passwords: nil,
-        extract_metadata: true
-      )
-        @extract_images = extract_images ? true : false
-        @passwords = if passwords.is_a?(Array)
-                       passwords.map(&:to_s)
-                     else
-                       (passwords ? [passwords.to_s] : nil)
-                     end
-        @extract_metadata = extract_metadata ? true : false
-      end
-
-      def to_h
-        {
-          extract_images: @extract_images,
-          passwords: @passwords,
-          extract_metadata: @extract_metadata
-        }.compact
-      end
-    end
-
-    # Image extraction configuration
-    #
-    # @example
-    #   image = ImageExtraction.new(extract_images: true, target_dpi: 300)
-    #
-    # @example With auto-adjust DPI
-    #   image = ImageExtraction.new(
-    #     extract_images: true,
-    #     auto_adjust_dpi: true,
-    #     min_dpi: 150,
-    #     max_dpi: 600
-    #   )
-    #
-    class ImageExtraction
-      attr_reader :extract_images, :target_dpi, :max_image_dimension,
-                  :auto_adjust_dpi, :min_dpi, :max_dpi
-
-      def initialize(
-        extract_images: true,
-        target_dpi: 300,
-        max_image_dimension: 2000,
-        auto_adjust_dpi: true,
-        min_dpi: 150,
-        max_dpi: 600
-      )
-        @extract_images = extract_images ? true : false
-        @target_dpi = target_dpi.to_i
-        @max_image_dimension = max_image_dimension.to_i
-        @auto_adjust_dpi = auto_adjust_dpi ? true : false
-        @min_dpi = min_dpi.to_i
-        @max_dpi = max_dpi.to_i
-      end
-
-      def to_h
-        {
-          extract_images: @extract_images,
-          target_dpi: @target_dpi,
-          max_image_dimension: @max_image_dimension,
-          auto_adjust_dpi: @auto_adjust_dpi,
-          min_dpi: @min_dpi,
-          max_dpi: @max_dpi
-        }
-      end
-    end
-
-    # Image preprocessing configuration for OCR
-    #
-    # @example Basic preprocessing
-    #   preprocessing = ImagePreprocessing.new(
-    #     binarization_method: "otsu",
-    #     denoise: true
-    #   )
-    #
-    # @example Advanced preprocessing
-    #   preprocessing = ImagePreprocessing.new(
-    #     target_dpi: 600,
-    #     auto_rotate: true,
-    #     deskew: true,
-    #     denoise: true,
-    #     contrast_enhance: true,
-    #     binarization_method: "sauvola",
-    #     invert_colors: false
-    #   )
-    #
-    class ImagePreprocessing
-      attr_reader :target_dpi, :auto_rotate, :deskew, :denoise,
-                  :contrast_enhance, :binarization_method, :invert_colors
-
-      def initialize(
-        target_dpi: 300,
-        auto_rotate: true,
-        deskew: true,
-        denoise: false,
-        contrast_enhance: true,
-        binarization_method: 'otsu',
-        invert_colors: false
-      )
-        @target_dpi = target_dpi.to_i
-        @auto_rotate = auto_rotate ? true : false
-        @deskew = deskew ? true : false
-        @denoise = denoise ? true : false
-        @contrast_enhance = contrast_enhance ? true : false
-        @binarization_method = binarization_method.to_s
-        @invert_colors = invert_colors ? true : false
-
-        valid_methods = %w[otsu sauvola adaptive]
-        return if valid_methods.include?(@binarization_method)
-
-        raise ArgumentError, "binarization_method must be one of: #{valid_methods.join(', ')}"
-      end
-
-      def to_h
-        {
-          target_dpi: @target_dpi,
-          auto_rotate: @auto_rotate,
-          deskew: @deskew,
-          denoise: @denoise,
-          contrast_enhance: @contrast_enhance,
-          binarization_method: @binarization_method,
-          invert_colors: @invert_colors
-        }
-      end
-    end
-
-    # Token reduction configuration
-    #
-    # @example Disable token reduction
-    #   token = TokenReduction.new(mode: "off")
-    #
-    # @example Light reduction
-    #   token = TokenReduction.new(mode: "light", preserve_important_words: true)
-    #
-    # @example Aggressive reduction
-    #   token = TokenReduction.new(mode: "aggressive", preserve_important_words: false)
-    #
-    class TokenReduction
-      attr_reader :mode, :preserve_important_words
-
-      def initialize(mode: 'off', preserve_important_words: true)
-        @mode = mode.to_s
-        @preserve_important_words = preserve_important_words ? true : false
-
-        valid_modes = %w[off light moderate aggressive maximum]
-        return if valid_modes.include?(@mode)
-
-        raise ArgumentError, "mode must be one of: #{valid_modes.join(', ')}"
-      end
-
-      def to_h
-        {
-          mode: @mode,
-          preserve_important_words: @preserve_important_words
-        }
-      end
-    end
-
-    class HtmlPreprocessing
-      attr_reader :enabled, :preset, :remove_navigation, :remove_forms
-
-      def initialize(enabled: nil, preset: nil, remove_navigation: nil, remove_forms: nil)
-        @enabled = boolean_or_nil(enabled)
-        @preset = preset&.to_sym
-        @remove_navigation = boolean_or_nil(remove_navigation)
-        @remove_forms = boolean_or_nil(remove_forms)
-      end
-
-      def to_h
-        {
-          enabled: @enabled,
-          preset: @preset,
-          remove_navigation: @remove_navigation,
-          remove_forms: @remove_forms
-        }.compact
-      end
-
-      private
-
-      def boolean_or_nil(value)
-        return nil if value.nil?
-
-        value ? true : false
-      end
-    end
-
-    class HtmlOptions
-      attr_reader :options
-
-      def initialize(**options)
-        normalized = options.transform_keys(&:to_sym)
-        symbol_keys = %i[
-          heading_style
-          code_block_style
-          highlight_style
-          list_indent_type
-          newline_style
-          whitespace_mode
-        ]
-        symbol_keys.each do |key|
-          normalized[key] = normalized[key]&.to_sym if normalized.key?(key)
-        end
-        if normalized[:preprocessing].is_a?(Hash)
-          normalized[:preprocessing] = HtmlPreprocessing.new(**normalized[:preprocessing])
-        end
-        @options = normalized
-      end
-
-      def to_h
-        @options.transform_values { |value| value.respond_to?(:to_h) ? value.to_h : value }
-      end
-    end
-
-    class KeywordYakeParams
-      attr_reader :window_size
-
-      def initialize(window_size: 2)
-        @window_size = window_size.to_i
-      end
-
-      def to_h
-        { window_size: @window_size }
-      end
-    end
-
-    class KeywordRakeParams
-      attr_reader :min_word_length, :max_words_per_phrase
-
-      def initialize(min_word_length: 1, max_words_per_phrase: 3)
-        @min_word_length = min_word_length.to_i
-        @max_words_per_phrase = max_words_per_phrase.to_i
-      end
-
-      def to_h
-        {
-          min_word_length: @min_word_length,
-          max_words_per_phrase: @max_words_per_phrase
-        }
-      end
-    end
-
-    class Keywords
-      attr_reader :algorithm, :max_keywords, :min_score, :ngram_range,
-                  :language, :yake_params, :rake_params
-
-      def initialize(
-        algorithm: nil,
-        max_keywords: nil,
-        min_score: nil,
-        ngram_range: nil,
-        language: nil,
-        yake_params: nil,
-        rake_params: nil
-      )
-        @algorithm = algorithm&.to_s
-        @max_keywords = max_keywords&.to_i
-        @min_score = min_score&.to_f
-        @ngram_range = ngram_range&.map(&:to_i)
-        @language = language&.to_s
-        @yake_params = normalize_nested(yake_params, KeywordYakeParams)
-        @rake_params = normalize_nested(rake_params, KeywordRakeParams)
-      end
-
-      def to_h
-        {
-          algorithm: @algorithm,
-          max_keywords: @max_keywords,
-          min_score: @min_score,
-          ngram_range: @ngram_range,
-          language: @language,
-          yake_params: @yake_params&.to_h,
-          rake_params: @rake_params&.to_h
-        }.compact
-      end
-
-      private
-
-      def normalize_nested(value, klass)
-        return nil if value.nil?
-        return value if value.is_a?(klass)
-        return klass.new(**value.transform_keys(&:to_sym)) if value.is_a?(Hash)
-
-        raise ArgumentError, "Expected #{klass}, Hash, or nil, got #{value.class}"
-      end
-    end
-
-    # Post-processor configuration
-    #
-    # @example Enable all post-processors
-    #   postprocessor = PostProcessor.new(enabled: true)
-    #
-    # @example Enable specific processors
-    #   postprocessor = PostProcessor.new(
-    #     enabled: true,
-    #     enabled_processors: ["quality", "formatting"]
-    #   )
-    #
-    # @example Disable specific processors
-    #   postprocessor = PostProcessor.new(
-    #     enabled: true,
-    #     disabled_processors: ["token_reduction"]
-    #   )
-    #
-    class PostProcessor
-      attr_reader :enabled, :enabled_processors, :disabled_processors
-
-      def initialize(
-        enabled: true,
-        enabled_processors: nil,
-        disabled_processors: nil
-      )
-        @enabled = enabled ? true : false
-        @enabled_processors = enabled_processors&.map(&:to_s)
-        @disabled_processors = disabled_processors&.map(&:to_s)
-      end
-
-      def to_h
-        {
-          enabled: @enabled,
-          enabled_processors: @enabled_processors,
-          disabled_processors: @disabled_processors
-        }.compact
-      end
-    end
-
-    # Main extraction configuration
-    #
-    # @example Basic usage
-    #   config = Extraction.new(use_cache: true, force_ocr: true)
-    #
-    # @example With OCR
-    #   ocr = Config::OCR.new(backend: "tesseract", language: "eng")
-    #   config = Extraction.new(ocr: ocr)
-    #
-    # @example With image extraction
-    #   image = Config::ImageExtraction.new(extract_images: true, target_dpi: 600)
-    #   config = Extraction.new(image_extraction: image)
-    #
-    # @example With preprocessing
-    #   preprocessing = Config::ImagePreprocessing.new(
-    #     binarization_method: "sauvola",
-    #     denoise: true
-    #   )
-    #   config = Extraction.new(image_preprocessing: preprocessing)
-    #
-    # @example With post-processing
-    #   postprocessor = Config::PostProcessor.new(
-    #     enabled: true,
-    #     enabled_processors: ["quality"]
-    #   )
-    #   config = Extraction.new(postprocessor: postprocessor)
-    #
-    # @example With all options
-    #   config = Extraction.new(
-    #     use_cache: true,
-    #     enable_quality_processing: true,
-    #     force_ocr: false,
-    #     ocr: Config::OCR.new(language: "deu"),
-    #     chunking: Config::Chunking.new(max_chars: 500),
-    #     language_detection: Config::LanguageDetection.new(enabled: true),
-    #     pdf_options: Config::PDF.new(extract_images: true, passwords: ["secret"]),
-    #     image_extraction: Config::ImageExtraction.new(target_dpi: 600),
-    #     image_preprocessing: Config::ImagePreprocessing.new(denoise: true),
-    #     postprocessor: Config::PostProcessor.new(enabled: true)
-    #   )
-    #
-    class Extraction
-      attr_reader :use_cache, :enable_quality_processing, :force_ocr,
-                  :ocr, :chunking, :language_detection, :pdf_options,
-                  :image_extraction, :image_preprocessing, :postprocessor,
-                  :token_reduction, :keywords, :html_options,
-                  :max_concurrent_extractions
-
-      # Load configuration from a file.
-      #
-      # Detects the file format from the extension (.toml, .yaml, .json)
-      # and loads the configuration accordingly.
-      #
-      # @param path [String] Path to the configuration file
-      # @return [Kreuzberg::Config::Extraction] Loaded configuration object
-      #
-      # @example Load from TOML
-      #   config = Kreuzberg::Config::Extraction.from_file("config.toml")
-      #
-      # @example Load from YAML
-      #   config = Kreuzberg::Config::Extraction.from_file("config.yaml")
-      #
-      def self.from_file(path)
-        hash = Kreuzberg._config_from_file_native(path)
-        # Convert string keys to symbols for keyword arguments
-        new(**hash.transform_keys(&:to_sym))
-      end
-
-      # Discover configuration file in current or parent directories.
-      #
-      # Searches for kreuzberg.toml, kreuzberg.yaml, or kreuzberg.json in the current
-      # directory and parent directories.
-      #
-      # @return [Kreuzberg::Config::Extraction, nil] Loaded configuration object or nil if not found
-      #
-      # @example
-      #   config = Kreuzberg::Config::Extraction.discover
-      #   if config
-      #     # Use discovered config
-      #   end
-      #
-      def self.discover
-        hash = Kreuzberg._config_discover_native
-        return nil if hash.nil?
-
-        # Convert string keys to symbols for keyword arguments
-        new(**hash.transform_keys(&:to_sym))
-      end
-
-      def initialize(
-        use_cache: true,
-        enable_quality_processing: false,
-        force_ocr: false,
-        ocr: nil,
-        chunking: nil,
-        language_detection: nil,
-        pdf_options: nil,
-        image_extraction: nil,
-        image_preprocessing: nil,
-        postprocessor: nil,
-        token_reduction: nil,
-        keywords: nil,
-        html_options: nil,
-        max_concurrent_extractions: nil
-      )
-        @use_cache = use_cache ? true : false
-        @enable_quality_processing = enable_quality_processing ? true : false
-        @force_ocr = force_ocr ? true : false
-        @ocr = normalize_config(ocr, OCR)
-        @chunking = normalize_config(chunking, Chunking)
-        @language_detection = normalize_config(language_detection, LanguageDetection)
-        @pdf_options = normalize_config(pdf_options, PDF)
-        @image_extraction = normalize_config(image_extraction, ImageExtraction)
-        @image_preprocessing = normalize_config(image_preprocessing, ImagePreprocessing)
-        @postprocessor = normalize_config(postprocessor, PostProcessor)
-        @token_reduction = normalize_config(token_reduction, TokenReduction)
-        @keywords = normalize_config(keywords, Keywords)
-        @html_options = normalize_config(html_options, HtmlOptions)
-        @max_concurrent_extractions = max_concurrent_extractions&.to_i
-      end
-
-      # rubocop:disable Metrics/PerceivedComplexity
-      def to_h
-        {
-          use_cache: @use_cache,
-          enable_quality_processing: @enable_quality_processing,
-          force_ocr: @force_ocr,
-          ocr: @ocr&.to_h,
-          chunking: @chunking&.to_h,
-          language_detection: @language_detection&.to_h,
-          pdf_options: @pdf_options&.to_h,
-          image_extraction: @image_extraction&.to_h,
-          image_preprocessing: @image_preprocessing&.to_h,
-          postprocessor: @postprocessor&.to_h,
-          token_reduction: @token_reduction&.to_h,
-          keywords: @keywords&.to_h,
-          html_options: @html_options&.to_h,
-          max_concurrent_extractions: @max_concurrent_extractions
-        }.compact
-      end
-      # rubocop:enable Metrics/PerceivedComplexity
-
-      private
-
-      def normalize_config(value, klass)
-        return nil if value.nil?
-        return value if value.is_a?(klass)
-        # Convert string keys to symbols for keyword arguments
-        return klass.new(**value.transform_keys(&:to_sym)) if value.is_a?(Hash)
-
-        raise ArgumentError, "Expected #{klass}, Hash, or nil, got #{value.class}"
-      end
-    end
-
-    # Backwards compatibility aliases
-    Ocr = OCR
-  end
-end
+# frozen_string_literal: true
+
+module Kreuzberg
+  module Config
+    # OCR configuration
+    #
+    # @example
+    #   ocr = OCR.new(backend: "tesseract", language: "eng")
+    #
+    class OCR
+      attr_reader :backend, :language, :tesseract_config
+
+      def initialize(
+        backend: 'tesseract',
+        language: 'eng',
+        tesseract_config: nil
+      )
+        @backend = backend.to_s
+        @language = language.to_s
+        @tesseract_config = normalize_tesseract_config(tesseract_config)
+      end
+
+      def to_h
+        {
+          backend: @backend,
+          language: @language,
+          tesseract_config: @tesseract_config&.to_h
+        }.compact
+      end
+
+      private
+
+      def normalize_tesseract_config(value)
+        return nil if value.nil?
+        return value if value.is_a?(Tesseract)
+        return Tesseract.new(**value.transform_keys(&:to_sym)) if value.is_a?(Hash)
+
+        raise ArgumentError, "Expected #{Tesseract}, Hash, or nil, got #{value.class}"
+      end
+    end
+
+    # Tesseract OCR engine configuration
+    class Tesseract
+      attr_reader :options
+
+      def initialize(**options)
+        @options = options.transform_keys(&:to_sym)
+        normalize_nested_preprocessing!
+      end
+
+      def to_h
+        @options.dup
+      end
+
+      private
+
+      def normalize_nested_preprocessing!
+        preprocessing = @options[:preprocessing]
+        return if preprocessing.nil?
+        return if preprocessing.is_a?(ImagePreprocessing)
+        return @options[:preprocessing] = ImagePreprocessing.new(**preprocessing.transform_keys(&:to_sym)) if
+          preprocessing.is_a?(Hash)
+
+        raise ArgumentError, "preprocessing must be #{ImagePreprocessing} or Hash"
+      end
+    end
+
+    # Chunking configuration
+    #
+    # @example
+    #   chunking = Chunking.new(max_chars: 1000, max_overlap: 200)
+    #
+    class Chunking
+      attr_reader :max_chars, :max_overlap, :preset, :embedding, :enabled
+
+      def initialize(
+        max_chars: nil,
+        max_overlap: nil,
+        preset: nil,
+        embedding: nil,
+        chunk_size: nil,
+        chunk_overlap: nil,
+        enabled: true
+      )
+        resolved_size = chunk_size || max_chars || 1000
+        resolved_overlap = chunk_overlap || max_overlap || 200
+
+        @max_chars = resolved_size.to_i
+        @max_overlap = resolved_overlap.to_i
+        @preset = preset&.to_s
+        @embedding = normalize_embedding(embedding)
+        @enabled = boolean_or_nil(enabled)
+      end
+
+      def to_h
+        config = {
+          max_chars: @max_chars,
+          max_overlap: @max_overlap,
+          preset: @preset,
+          embedding: @embedding&.to_h
+        }.compact
+        # @type var config: Hash[Symbol, untyped]
+        config[:enabled] = @enabled unless @enabled.nil?
+        config
+      end
+
+      private
+
+      def normalize_embedding(value)
+        return nil if value.nil?
+        return value if value.is_a?(Embedding)
+        return Embedding.new(**value.transform_keys(&:to_sym)) if value.is_a?(Hash)
+
+        raise ArgumentError, "Expected #{Embedding}, Hash, or nil, got #{value.class}"
+      end
+
+      def boolean_or_nil(value)
+        return nil if value.nil?
+
+        value ? true : false
+      end
+    end
+
+    # Embedding model configuration for document chunking
+    class Embedding
+      attr_reader :model, :normalize, :batch_size, :show_download_progress, :cache_dir
+
+      def initialize(
+        model: { type: :preset, name: 'balanced' },
+        normalize: true,
+        batch_size: 32,
+        show_download_progress: false,
+        cache_dir: nil
+      )
+        @model = normalize_model(model)
+        @normalize = boolean_or_nil(normalize)
+        @batch_size = batch_size&.to_i
+        @show_download_progress = boolean_or_nil(show_download_progress)
+        @cache_dir = cache_dir&.to_s
+      end
+
+      def to_h
+        {
+          model: @model,
+          normalize: @normalize,
+          batch_size: @batch_size,
+          show_download_progress: @show_download_progress,
+          cache_dir: @cache_dir
+        }.compact
+      end
+
+      private
+
+      def normalize_model(model)
+        normalized = if model.respond_to?(:to_h)
+                       model.to_h
+                     else
+                       model
+                     end
+        raise ArgumentError, 'model must be a Hash describing the embedding model' unless normalized.is_a?(Hash)
+
+        normalized.transform_keys(&:to_sym)
+      end
+
+      def boolean_or_nil(value)
+        return nil if value.nil?
+
+        value ? true : false
+      end
+    end
+
+    # Language detection configuration
+    #
+    # @example
+    #   lang = LanguageDetection.new(enabled: true, min_confidence: 0.8)
+    #
+    class LanguageDetection
+      attr_reader :enabled, :min_confidence, :detect_multiple
+
+      def initialize(enabled: false, min_confidence: 0.5, detect_multiple: false)
+        @enabled = enabled ? true : false
+        @min_confidence = min_confidence.to_f
+        @detect_multiple = detect_multiple ? true : false
+      end
+
+      def to_h
+        {
+          enabled: @enabled,
+          min_confidence: @min_confidence,
+          detect_multiple: @detect_multiple
+        }
+      end
+    end
+
+    # PDF-specific options
+    #
+    # @example
+    #   pdf = PDF.new(extract_images: true, passwords: ["secret", "backup"])
+    #
+    class PDF
+      attr_reader :extract_images, :passwords, :extract_metadata
+
+      def initialize(
+        extract_images: false,
+        passwords: nil,
+        extract_metadata: true
+      )
+        @extract_images = extract_images ? true : false
+        @passwords = if passwords.is_a?(Array)
+                       passwords.map(&:to_s)
+                     else
+                       (passwords ? [passwords.to_s] : nil)
+                     end
+        @extract_metadata = extract_metadata ? true : false
+      end
+
+      def to_h
+        {
+          extract_images: @extract_images,
+          passwords: @passwords,
+          extract_metadata: @extract_metadata
+        }.compact
+      end
+    end
+
+    # Image extraction configuration
+    #
+    # @example
+    #   image = ImageExtraction.new(extract_images: true, target_dpi: 300)
+    #
+    # @example With auto-adjust DPI
+    #   image = ImageExtraction.new(
+    #     extract_images: true,
+    #     auto_adjust_dpi: true,
+    #     min_dpi: 150,
+    #     max_dpi: 600
+    #   )
+    #
+    class ImageExtraction
+      attr_reader :extract_images, :target_dpi, :max_image_dimension,
+                  :auto_adjust_dpi, :min_dpi, :max_dpi
+
+      def initialize(
+        extract_images: true,
+        target_dpi: 300,
+        max_image_dimension: 2000,
+        auto_adjust_dpi: true,
+        min_dpi: 150,
+        max_dpi: 600
+      )
+        @extract_images = extract_images ? true : false
+        @target_dpi = target_dpi.to_i
+        @max_image_dimension = max_image_dimension.to_i
+        @auto_adjust_dpi = auto_adjust_dpi ? true : false
+        @min_dpi = min_dpi.to_i
+        @max_dpi = max_dpi.to_i
+      end
+
+      def to_h
+        {
+          extract_images: @extract_images,
+          target_dpi: @target_dpi,
+          max_image_dimension: @max_image_dimension,
+          auto_adjust_dpi: @auto_adjust_dpi,
+          min_dpi: @min_dpi,
+          max_dpi: @max_dpi
+        }
+      end
+    end
+
+    # Image preprocessing configuration for OCR
+    #
+    # @example Basic preprocessing
+    #   preprocessing = ImagePreprocessing.new(
+    #     binarization_method: "otsu",
+    #     denoise: true
+    #   )
+    #
+    # @example Advanced preprocessing
+    #   preprocessing = ImagePreprocessing.new(
+    #     target_dpi: 600,
+    #     auto_rotate: true,
+    #     deskew: true,
+    #     denoise: true,
+    #     contrast_enhance: true,
+    #     binarization_method: "sauvola",
+    #     invert_colors: false
+    #   )
+    #
+    class ImagePreprocessing
+      attr_reader :target_dpi, :auto_rotate, :deskew, :denoise,
+                  :contrast_enhance, :binarization_method, :invert_colors
+
+      def initialize(
+        target_dpi: 300,
+        auto_rotate: true,
+        deskew: true,
+        denoise: false,
+        contrast_enhance: true,
+        binarization_method: 'otsu',
+        invert_colors: false
+      )
+        @target_dpi = target_dpi.to_i
+        @auto_rotate = auto_rotate ? true : false
+        @deskew = deskew ? true : false
+        @denoise = denoise ? true : false
+        @contrast_enhance = contrast_enhance ? true : false
+        @binarization_method = binarization_method.to_s
+        @invert_colors = invert_colors ? true : false
+
+        valid_methods = %w[otsu sauvola adaptive]
+        return if valid_methods.include?(@binarization_method)
+
+        raise ArgumentError, "binarization_method must be one of: #{valid_methods.join(', ')}"
+      end
+
+      def to_h
+        {
+          target_dpi: @target_dpi,
+          auto_rotate: @auto_rotate,
+          deskew: @deskew,
+          denoise: @denoise,
+          contrast_enhance: @contrast_enhance,
+          binarization_method: @binarization_method,
+          invert_colors: @invert_colors
+        }
+      end
+    end
+
+    # Token reduction configuration
+    #
+    # @example Disable token reduction
+    #   token = TokenReduction.new(mode: "off")
+    #
+    # @example Light reduction
+    #   token = TokenReduction.new(mode: "light", preserve_important_words: true)
+    #
+    # @example Aggressive reduction
+    #   token = TokenReduction.new(mode: "aggressive", preserve_important_words: false)
+    #
+    class TokenReduction
+      attr_reader :mode, :preserve_important_words
+
+      def initialize(mode: 'off', preserve_important_words: true)
+        @mode = mode.to_s
+        @preserve_important_words = preserve_important_words ? true : false
+
+        valid_modes = %w[off light moderate aggressive maximum]
+        return if valid_modes.include?(@mode)
+
+        raise ArgumentError, "mode must be one of: #{valid_modes.join(', ')}"
+      end
+
+      def to_h
+        {
+          mode: @mode,
+          preserve_important_words: @preserve_important_words
+        }
+      end
+    end
+
+    # HTML preprocessing configuration for content extraction
+    class HtmlPreprocessing
+      attr_reader :enabled, :preset, :remove_navigation, :remove_forms
+
+      def initialize(enabled: nil, preset: nil, remove_navigation: nil, remove_forms: nil)
+        @enabled = boolean_or_nil(enabled)
+        @preset = preset&.to_sym
+        @remove_navigation = boolean_or_nil(remove_navigation)
+        @remove_forms = boolean_or_nil(remove_forms)
+      end
+
+      def to_h
+        {
+          enabled: @enabled,
+          preset: @preset,
+          remove_navigation: @remove_navigation,
+          remove_forms: @remove_forms
+        }.compact
+      end
+
+      private
+
+      def boolean_or_nil(value)
+        return nil if value.nil?
+
+        value ? true : false
+      end
+    end
+
+    # HTML rendering options for document conversion
+    class HtmlOptions
+      attr_reader :options
+
+      def initialize(**options)
+        normalized = options.transform_keys(&:to_sym)
+        symbol_keys = %i[
+          heading_style
+          code_block_style
+          highlight_style
+          list_indent_type
+          newline_style
+          whitespace_mode
+        ]
+        symbol_keys.each do |key|
+          normalized[key] = normalized[key]&.to_sym if normalized.key?(key)
+        end
+        if normalized[:preprocessing].is_a?(Hash)
+          normalized[:preprocessing] = HtmlPreprocessing.new(**normalized[:preprocessing])
+        end
+        @options = normalized
+      end
+
+      def to_h
+        @options.transform_values { |value| value.respond_to?(:to_h) ? value.to_h : value }
+      end
+    end
+
+    # YAKE keyword extraction parameters
+    class KeywordYakeParams
+      attr_reader :window_size
+
+      def initialize(window_size: 2)
+        @window_size = window_size.to_i
+      end
+
+      def to_h
+        { window_size: @window_size }
+      end
+    end
+
+    # RAKE keyword extraction parameters
+    class KeywordRakeParams
+      attr_reader :min_word_length, :max_words_per_phrase
+
+      def initialize(min_word_length: 1, max_words_per_phrase: 3)
+        @min_word_length = min_word_length.to_i
+        @max_words_per_phrase = max_words_per_phrase.to_i
+      end
+
+      def to_h
+        {
+          min_word_length: @min_word_length,
+          max_words_per_phrase: @max_words_per_phrase
+        }
+      end
+    end
+
+    # Keyword extraction configuration for document analysis
+    class Keywords
+      attr_reader :algorithm, :max_keywords, :min_score, :ngram_range,
+                  :language, :yake_params, :rake_params
+
+      def initialize(
+        algorithm: nil,
+        max_keywords: nil,
+        min_score: nil,
+        ngram_range: nil,
+        language: nil,
+        yake_params: nil,
+        rake_params: nil
+      )
+        @algorithm = algorithm&.to_s
+        @max_keywords = max_keywords&.to_i
+        @min_score = min_score&.to_f
+        @ngram_range = ngram_range&.map(&:to_i)
+        @language = language&.to_s
+        @yake_params = normalize_nested(yake_params, KeywordYakeParams)
+        @rake_params = normalize_nested(rake_params, KeywordRakeParams)
+      end
+
+      def to_h
+        {
+          algorithm: @algorithm,
+          max_keywords: @max_keywords,
+          min_score: @min_score,
+          ngram_range: @ngram_range,
+          language: @language,
+          yake_params: @yake_params&.to_h,
+          rake_params: @rake_params&.to_h
+        }.compact
+      end
+
+      private
+
+      def normalize_nested(value, klass)
+        return nil if value.nil?
+        return value if value.is_a?(klass)
+        return klass.new(**value.transform_keys(&:to_sym)) if value.is_a?(Hash)
+
+        raise ArgumentError, "Expected #{klass}, Hash, or nil, got #{value.class}"
+      end
+    end
+
+    # Post-processor configuration
+    #
+    # @example Enable all post-processors
+    #   postprocessor = PostProcessor.new(enabled: true)
+    #
+    # @example Enable specific processors
+    #   postprocessor = PostProcessor.new(
+    #     enabled: true,
+    #     enabled_processors: ["quality", "formatting"]
+    #   )
+    #
+    # @example Disable specific processors
+    #   postprocessor = PostProcessor.new(
+    #     enabled: true,
+    #     disabled_processors: ["token_reduction"]
+    #   )
+    #
+    class PostProcessor
+      attr_reader :enabled, :enabled_processors, :disabled_processors
+
+      def initialize(
+        enabled: true,
+        enabled_processors: nil,
+        disabled_processors: nil
+      )
+        @enabled = enabled ? true : false
+        @enabled_processors = enabled_processors&.map(&:to_s)
+        @disabled_processors = disabled_processors&.map(&:to_s)
+      end
+
+      def to_h
+        {
+          enabled: @enabled,
+          enabled_processors: @enabled_processors,
+          disabled_processors: @disabled_processors
+        }.compact
+      end
+    end
+
+    # Main extraction configuration
+    #
+    # @example Basic usage
+    #   config = Extraction.new(use_cache: true, force_ocr: true)
+    #
+    # @example With OCR
+    #   ocr = Config::OCR.new(backend: "tesseract", language: "eng")
+    #   config = Extraction.new(ocr: ocr)
+    #
+    # @example With image extraction
+    #   image = Config::ImageExtraction.new(extract_images: true, target_dpi: 600)
+    #   config = Extraction.new(image_extraction: image)
+    #
+    # @example With preprocessing
+    #   preprocessing = Config::ImagePreprocessing.new(
+    #     binarization_method: "sauvola",
+    #     denoise: true
+    #   )
+    #   config = Extraction.new(image_preprocessing: preprocessing)
+    #
+    # @example With post-processing
+    #   postprocessor = Config::PostProcessor.new(
+    #     enabled: true,
+    #     enabled_processors: ["quality"]
+    #   )
+    #   config = Extraction.new(postprocessor: postprocessor)
+    #
+    # @example With all options
+    #   config = Extraction.new(
+    #     use_cache: true,
+    #     enable_quality_processing: true,
+    #     force_ocr: false,
+    #     ocr: Config::OCR.new(language: "deu"),
+    #     chunking: Config::Chunking.new(max_chars: 500),
+    #     language_detection: Config::LanguageDetection.new(enabled: true),
+    #     pdf_options: Config::PDF.new(extract_images: true, passwords: ["secret"]),
+    #     image_extraction: Config::ImageExtraction.new(target_dpi: 600),
+    #     image_preprocessing: Config::ImagePreprocessing.new(denoise: true),
+    #     postprocessor: Config::PostProcessor.new(enabled: true)
+    #   )
+    #
+    class Extraction
+      attr_reader :use_cache, :enable_quality_processing, :force_ocr,
+                  :ocr, :chunking, :language_detection, :pdf_options,
+                  :image_extraction, :image_preprocessing, :postprocessor,
+                  :token_reduction, :keywords, :html_options,
+                  :max_concurrent_extractions
+
+      # Load configuration from a file.
+      #
+      # Detects the file format from the extension (.toml, .yaml, .json)
+      # and loads the configuration accordingly.
+      #
+      # @param path [String] Path to the configuration file
+      # @return [Kreuzberg::Config::Extraction] Loaded configuration object
+      #
+      # @example Load from TOML
+      #   config = Kreuzberg::Config::Extraction.from_file("config.toml")
+      #
+      # @example Load from YAML
+      #   config = Kreuzberg::Config::Extraction.from_file("config.yaml")
+      #
+      def self.from_file(path)
+        hash = Kreuzberg._config_from_file_native(path)
+        # Convert string keys to symbols for keyword arguments
+        new(**hash.transform_keys(&:to_sym))
+      end
+
+      # Discover configuration file in current or parent directories.
+      #
+      # Searches for kreuzberg.toml, kreuzberg.yaml, or kreuzberg.json in the current
+      # directory and parent directories.
+      #
+      # @return [Kreuzberg::Config::Extraction, nil] Loaded configuration object or nil if not found
+      #
+      # @example
+      #   config = Kreuzberg::Config::Extraction.discover
+      #   if config
+      #     # Use discovered config
+      #   end
+      #
+      def self.discover
+        hash = Kreuzberg._config_discover_native
+        return nil if hash.nil?
+
+        # Convert string keys to symbols for keyword arguments
+        new(**hash.transform_keys(&:to_sym))
+      end
+
+      def initialize(
+        use_cache: true,
+        enable_quality_processing: false,
+        force_ocr: false,
+        ocr: nil,
+        chunking: nil,
+        language_detection: nil,
+        pdf_options: nil,
+        image_extraction: nil,
+        image_preprocessing: nil,
+        postprocessor: nil,
+        token_reduction: nil,
+        keywords: nil,
+        html_options: nil,
+        max_concurrent_extractions: nil
+      )
+        @use_cache = use_cache ? true : false
+        @enable_quality_processing = enable_quality_processing ? true : false
+        @force_ocr = force_ocr ? true : false
+        @ocr = normalize_config(ocr, OCR)
+        @chunking = normalize_config(chunking, Chunking)
+        @language_detection = normalize_config(language_detection, LanguageDetection)
+        @pdf_options = normalize_config(pdf_options, PDF)
+        @image_extraction = normalize_config(image_extraction, ImageExtraction)
+        @image_preprocessing = normalize_config(image_preprocessing, ImagePreprocessing)
+        @postprocessor = normalize_config(postprocessor, PostProcessor)
+        @token_reduction = normalize_config(token_reduction, TokenReduction)
+        @keywords = normalize_config(keywords, Keywords)
+        @html_options = normalize_config(html_options, HtmlOptions)
+        @max_concurrent_extractions = max_concurrent_extractions&.to_i
+      end
+
+      # rubocop:disable Metrics/CyclomaticComplexity
+      def to_h
+        {
+          use_cache: @use_cache,
+          enable_quality_processing: @enable_quality_processing,
+          force_ocr: @force_ocr,
+          ocr: @ocr&.to_h,
+          chunking: @chunking&.to_h,
+          language_detection: @language_detection&.to_h,
+          pdf_options: @pdf_options&.to_h,
+          image_extraction: @image_extraction&.to_h,
+          image_preprocessing: @image_preprocessing&.to_h,
+          postprocessor: @postprocessor&.to_h,
+          token_reduction: @token_reduction&.to_h,
+          keywords: @keywords&.to_h,
+          html_options: @html_options&.to_h,
+          max_concurrent_extractions: @max_concurrent_extractions
+        }.compact
+      end
+      # rubocop:enable Metrics/CyclomaticComplexity
+
+      private
+
+      def normalize_config(value, klass)
+        return nil if value.nil?
+        return value if value.is_a?(klass)
+        # Convert string keys to symbols for keyword arguments
+        return klass.new(**value.transform_keys(&:to_sym)) if value.is_a?(Hash)
+
+        raise ArgumentError, "Expected #{klass}, Hash, or nil, got #{value.class}"
+      end
+    end
+
+    # Backwards compatibility aliases
+    Ocr = OCR
+  end
+end