npm - @kreuzberg/node - Versions diffs - 4.0.0-rc.15 → 4.0.0-rc.16 - Mend

@kreuzberg/node 4.0.0-rc.15 → 4.0.0-rc.16

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (12) hide show

package/dist/types.js.map CHANGED Viewed

	@@ -1 +1 @@
1	- {"version":3,"sources":["../typescript/types.ts"],"sourcesContent":["/*\n Type definitions for Kreuzberg extraction results.\n \n These types mirror the strongly-typed Rust metadata structures,\n * providing type safety for TypeScript users.\n /\n\n// ============================================================================\n// ============================================================================\n\n/\n Tesseract OCR engine configuration options.\n \n @example\n * ```typescript\n * const config: TesseractConfig = {\n * psm: 6,\n * enableTableDetection: true,\n * tesseditCharWhitelist: '0123456789'\n * };\n * ```\n /\nexport interface TesseractConfig {\n\t/\n\t Page Segmentation Mode (0-13). Controls how Tesseract segments and recognizes text.\n\t * Common values: 3 (auto), 6 (single uniform block), 11 (sparse text).\n\t * Default: 3 (auto layout analysis).\n\t /\n\tpsm?: number;\n\n\t/\n\t Enable table detection during OCR processing.\n\t * When true, Tesseract attempts to preserve table structure in the output.\n\t * Default: false.\n\t /\n\tenableTableDetection?: boolean;\n\n\t/\n\t Whitelist of characters Tesseract should recognize.\n\t * Only these characters will be returned by the OCR engine.\n\t * Use empty string to allow all characters. Useful for constraining output to digits,\n\t * specific alphabets, or other character sets.\n\t * Default: null (recognize all).\n\t /\n\ttesseditCharWhitelist?: string;\n}\n\n/\n OCR (Optical Character Recognition) configuration.\n \n Controls which OCR engine to use and how it processes images.\n /\nexport interface OcrConfig {\n\t/* OCR backend name (e.g., 'tesseract', 'paddleocr', 'easyocr'). Required. /\n\tbackend: string;\n\n\t/* ISO 639-1/3 language code(s) for OCR (e.g., 'eng', 'fra', 'deu'). Default: 'eng'. /\n\tlanguage?: string;\n\n\t/* Tesseract engine-specific configuration options. Only used when backend is 'tesseract'. /\n\ttesseractConfig?: TesseractConfig;\n}\n\n/\n Document chunking configuration for splitting large documents.\n \n Breaks large documents into smaller, manageable chunks while preserving context.\n * Useful for RAG (Retrieval Augmented Generation) and vector database indexing.\n /\nexport interface ChunkingConfig {\n\t/* Maximum characters per chunk. Default: 4096. /\n\tmaxChars?: number;\n\n\t/* Maximum overlapping characters between consecutive chunks for context preservation. Default: 512. /\n\tmaxOverlap?: number;\n\n\t/\n\t Alternative to maxChars: chunk size using different unit.\n\t * Mutually exclusive with maxChars.\n\t /\n\tchunkSize?: number;\n\n\t/\n\t Alternative to maxOverlap: overlap amount using different unit.\n\t * Mutually exclusive with maxOverlap.\n\t /\n\tchunkOverlap?: number;\n\n\t/\n\t Named preset configuration (e.g., 'default', 'aggressive', 'minimal').\n\t * Uses preset values if neither maxChars nor chunkSize is specified.\n\t /\n\tpreset?: string;\n\n\t/* Embedding configuration for generating vector embeddings for each chunk. /\n\tembedding?: Record<string, unknown>;\n\n\t/* Enable or disable chunking. Default: true when chunking config is provided. /\n\tenabled?: boolean;\n}\n\n/\n Language detection configuration.\n \n Automatically detects the language(s) of extracted content.\n /\nexport interface LanguageDetectionConfig {\n\t/* Enable automatic language detection. Default: true. /\n\tenabled?: boolean;\n\n\t/* Minimum confidence score (0.0-1.0) for language detection. Default: 0.5. /\n\tminConfidence?: number;\n\n\t/* Detect multiple languages in the same document. Default: false. /\n\tdetectMultiple?: boolean;\n}\n\n/\n Token reduction configuration for optimizing token usage.\n \n Reduces the number of tokens in extracted content while preserving meaning.\n * Useful for reducing costs in LLM pipelines.\n /\nexport interface TokenReductionConfig {\n\t/* Reduction mode: 'aggressive' or 'conservative'. Default: 'conservative'. /\n\tmode?: string;\n\n\t/* Preserve tokens for semantically important words even in aggressive mode. Default: true. /\n\tpreserveImportantWords?: boolean;\n}\n\n/\n PDF-specific extraction configuration.\n \n Controls how PDF documents are processed.\n /\nexport interface PdfConfig {\n\t/* Extract images from PDF pages. Default: true. /\n\textractImages?: boolean;\n\n\t/* List of passwords to try for password-protected PDFs. /\n\tpasswords?: string[];\n\n\t/* Extract document metadata (title, author, creation date, etc.). Default: true. /\n\textractMetadata?: boolean;\n}\n\n/\n Image extraction and processing configuration.\n \n Controls how images are extracted and optimized from documents.\n /\nexport interface ImageExtractionConfig {\n\t/* Enable image extraction from documents. Default: true. /\n\textractImages?: boolean;\n\n\t/* Target DPI (dots per inch) for extracted images. Higher DPI = better quality but larger files. Default: 150. /\n\ttargetDpi?: number;\n\n\t/* Maximum image dimension (width or height) in pixels. Images larger than this are downscaled. Default: 2000. /\n\tmaxImageDimension?: number;\n\n\t/* Automatically adjust DPI based on image content and quality. Default: true. /\n\tautoAdjustDpi?: boolean;\n\n\t/* Minimum DPI to maintain for image quality. Default: 72. /\n\tminDpi?: number;\n\n\t/* Maximum DPI to avoid excessive file sizes. Default: 300. /\n\tmaxDpi?: number;\n}\n\n/\n Post-processor configuration for modifying extracted content.\n \n Post-processors allow customization and cleanup of extraction results\n * without failing the extraction if they encounter errors.\n /\nexport interface PostProcessorConfig {\n\t/* Enable or disable post-processing entirely. Default: true. /\n\tenabled?: boolean;\n\n\t/* List of processor names to enable (allowlist). When set, only these are used. /\n\tenabledProcessors?: string[];\n\n\t/* List of processor names to disable (denylist). These are skipped. /\n\tdisabledProcessors?: string[];\n}\n\n/\n HTML preprocessing options.\n \n Cleans HTML content before conversion to Markdown.\n /\nexport interface HtmlPreprocessingOptions {\n\t/* Enable HTML preprocessing. Default: true. /\n\tenabled?: boolean;\n\n\t/* Preset cleanup level: 'minimal' (light), 'standard' (balanced), 'aggressive' (heavy). Default: 'standard'. /\n\tpreset?: \"minimal\" \| \"standard\" \| \"aggressive\";\n\n\t/* Remove navigation menus and headers. Default: true. /\n\tremoveNavigation?: boolean;\n\n\t/* Remove form elements. Default: true. /\n\tremoveForms?: boolean;\n}\n\n/\n HTML to Markdown conversion configuration options.\n \n Controls how HTML content is converted to Markdown format, including formatting,\n * escaping, and special handling for various HTML elements.\n /\nexport interface HtmlConversionOptions {\n\t/* Heading style conversion: \"atx\" (# style), \"underlined\" (underline style), or \"atx_closed\" (# style closed). Default: \"atx\". /\n\theadingStyle?: \"atx\" \| \"underlined\" \| \"atx_closed\";\n\n\t/* List indentation type: \"spaces\" or \"tabs\". Default: \"spaces\". /\n\tlistIndentType?: \"spaces\" \| \"tabs\";\n\n\t/* Number of spaces/tabs per list indent level. Default: 4. /\n\tlistIndentWidth?: number;\n\n\t/* Bullet characters for unordered lists (e.g., '', '-', '+'). Default: ''. /\n\tbullets?: string;\n\n\t/* Markdown symbol for strong/bold emphasis: '' or '__'. Default: ''. /\n\tstrongEmSymbol?: string;\n\n\t/* Escape asterisks () in text to prevent accidental formatting. Default: false. /\n\tescapeAsterisks?: boolean;\n\n\t/** Escape underscores (_) in text to prevent accidental formatting. Default: false. /\n\tescapeUnderscores?: boolean;\n\n\t/* Escape miscellaneous special characters. Default: false. /\n\tescapeMisc?: boolean;\n\n\t/* Escape ASCII control characters. Default: false. /\n\tescapeAscii?: boolean;\n\n\t/* Default code language for syntax highlighting in code blocks (e.g., 'javascript'). Default: null. /\n\tcodeLanguage?: string;\n\n\t/* Convert HTML links to Markdown autolinks format ([text](url)). Default: true. /\n\tautolinks?: boolean;\n\n\t/* Use the HTML title element as default for links when no text is available. Default: false. /\n\tdefaultTitle?: boolean;\n\n\t/* Insert <br> tags in Markdown tables. Default: false. /\n\tbrInTables?: boolean;\n\n\t/* Use HOCR spatial table format for better table structure preservation. Default: false. /\n\thocrSpatialTables?: boolean;\n\n\t/* Highlight style for marked/highlighted text: \"double_equal\" (==text==), \"html\" (<mark>), \"bold\" (text), or \"none\". Default: \"none\". /\n\thighlightStyle?: \"double_equal\" \| \"html\" \| \"bold\" \| \"none\";\n\n\t/* Extract metadata from HTML (title, meta tags, etc.). Default: false. /\n\textractMetadata?: boolean;\n\n\t/* Whitespace handling: \"normalized\" (collapse whitespace) or \"strict\" (preserve all whitespace). Default: \"normalized\". /\n\twhitespaceMode?: \"normalized\" \| \"strict\";\n\n\t/* Remove newlines from output (convert to single line). Default: false. /\n\tstripNewlines?: boolean;\n\n\t/* Enable line wrapping at specified width. Default: true. /\n\twrap?: boolean;\n\n\t/* Maximum line width when wrapping is enabled. Default: 80. /\n\twrapWidth?: number;\n\n\t/* Convert as inline Markdown instead of block elements. Default: false. /\n\tconvertAsInline?: boolean;\n\n\t/* Markdown symbol for subscript text (e.g., '~' for ~text~). Default: '~'. /\n\tsubSymbol?: string;\n\n\t/* Markdown symbol for superscript text (e.g., '^' for ^text^). Default: '^'. /\n\tsupSymbol?: string;\n\n\t/* Newline style in output: \"spaces\" (two spaces + newline) or \"backslash\" (backslash + newline). Default: \"spaces\". /\n\tnewlineStyle?: \"spaces\" \| \"backslash\";\n\n\t/* Code block style: \"indented\" (4-space indent), \"backticks\" (```), or \"tildes\" (~~~). Default: \"backticks\". /\n\tcodeBlockStyle?: \"indented\" \| \"backticks\" \| \"tildes\";\n\n\t/* List of HTML tag names to keep as inline images (don't convert). Default: []. /\n\tkeepInlineImagesIn?: string[];\n\n\t/* Character encoding for output (e.g., 'utf-8', 'ascii'). Default: 'utf-8'. /\n\tencoding?: string;\n\n\t/* Enable debug mode for detailed conversion logging. Default: false. /\n\tdebug?: boolean;\n\n\t/* List of HTML tag names to remove entirely from output. Default: []. /\n\tstripTags?: string[];\n\n\t/* List of HTML tag names to preserve in output (don't convert to Markdown). Default: []. /\n\tpreserveTags?: string[];\n\n\t/* HTML preprocessing options for cleaning HTML before conversion. /\n\tpreprocessing?: HtmlPreprocessingOptions;\n}\n\n/* Keyword extraction algorithm type. /\nexport type KeywordAlgorithm = \"yake\" \| \"rake\";\n\n/\n YAKE (Yet Another Keyword Extractor) algorithm configuration.\n \n YAKE is an unsupervised keyword extraction method that doesn't require training data.\n /\nexport interface YakeParams {\n\t/* Window size for co-occurrence analysis (number of words to consider). Default: 3. /\n\twindowSize?: number;\n}\n\n/\n RAKE (Rapid Automatic Keyword Extraction) algorithm configuration.\n \n RAKE extracts keywords based on word co-occurrence and statistical measures.\n /\nexport interface RakeParams {\n\t/* Minimum word length to consider as keyword. Default: 3. /\n\tminWordLength?: number;\n\n\t/* Maximum number of words per keyword phrase. Default: 3. /\n\tmaxWordsPerPhrase?: number;\n}\n\n/\n Keyword extraction configuration.\n \n Extracts important keywords/phrases from document content using YAKE or RAKE algorithms.\n /\nexport interface KeywordConfig {\n\t/* Extraction algorithm: \"yake\" or \"rake\". Default: \"yake\". /\n\talgorithm?: KeywordAlgorithm;\n\n\t/* Maximum number of keywords to extract. Default: 10. /\n\tmaxKeywords?: number;\n\n\t/* Minimum relevance score (0.0-1.0) for keywords. Keywords below this are filtered out. Default: 0.1. /\n\tminScore?: number;\n\n\t/* N-gram range: [min_length, max_length] for phrase keywords (e.g., [1, 3] for 1-3 word phrases). Default: [1, 3]. /\n\tngramRange?: [number, number];\n\n\t/* Language for keyword extraction (e.g., 'en', 'de', 'fr'). Default: 'en'. /\n\tlanguage?: string;\n\n\t/* YAKE algorithm-specific parameters. Only used when algorithm is \"yake\". /\n\tyakeParams?: YakeParams;\n\n\t/* RAKE algorithm-specific parameters. Only used when algorithm is \"rake\". /\n\trakeParams?: RakeParams;\n}\n\n/\n Page tracking and extraction configuration.\n \n Controls how pages/slides/sheets are extracted and tracked in the document.\n * Page range information in chunk metadata (first_page/last_page) is automatically\n * enabled when page boundaries are available and chunking is configured.\n /\nexport interface PageConfig {\n\t/* Extract pages as separate array (ExtractionResult.pages) /\n\textractPages?: boolean;\n\t/* Insert page markers in main content string /\n\tinsertPageMarkers?: boolean;\n\t/* Page marker format (use {page_num} placeholder) /\n\tmarkerFormat?: string;\n}\n\n/\n Main extraction configuration interface.\n \n Combines all sub-configurations for document extraction, OCR, chunking, post-processing, etc.\n * All fields are optional and use sensible defaults.\n /\nexport interface ExtractionConfig {\n\t/* Enable caching of extraction results for identical inputs. Default: true. /\n\tuseCache?: boolean;\n\n\t/* Enable quality processing filters to improve extraction reliability. Default: false. /\n\tenableQualityProcessing?: boolean;\n\n\t/* OCR configuration for text extraction from images. Only used when document contains images or forceOcr is true. /\n\tocr?: OcrConfig;\n\n\t/* Force OCR processing even for documents with selectable text. Useful for scanned documents. Default: false. /\n\tforceOcr?: boolean;\n\n\t/* Chunking configuration for splitting documents into smaller pieces for RAG or vector DB. /\n\tchunking?: ChunkingConfig;\n\n\t/* Image extraction and optimization configuration. /\n\timages?: ImageExtractionConfig;\n\n\t/* PDF-specific extraction options (passwords, metadata, etc.). /\n\tpdfOptions?: PdfConfig;\n\n\t/* Token reduction configuration for optimizing token usage in LLM pipelines. /\n\ttokenReduction?: TokenReductionConfig;\n\n\t/* Language detection configuration for automatic language identification. /\n\tlanguageDetection?: LanguageDetectionConfig;\n\n\t/* Post-processor configuration for customizing extraction results. /\n\tpostprocessor?: PostProcessorConfig;\n\n\t/* HTML to Markdown conversion options for HTML content. /\n\thtmlOptions?: HtmlConversionOptions;\n\n\t/* Keyword extraction configuration for extracting important phrases. /\n\tkeywords?: KeywordConfig;\n\n\t/* Page tracking and extraction configuration for multi-page documents. /\n\tpages?: PageConfig;\n\n\t/* Maximum number of concurrent extractions in batch operations. Default: 4. /\n\tmaxConcurrentExtractions?: number;\n}\n\n/\n Extracted table data from document.\n \n Contains both cell data and Markdown representation for easy display and processing.\n /\nexport interface Table {\n\t/* 2D array of cell contents (rows × columns) /\n\tcells: string[][];\n\n\t/* Markdown representation of the table for display or parsing /\n\tmarkdown: string;\n\n\t/* Page number where this table was found (1-indexed) /\n\tpageNumber: number;\n}\n\nexport interface ExcelMetadata {\n\tsheetCount?: number;\n\tsheetNames?: string[];\n}\n\nexport interface EmailMetadata {\n\tfromEmail?: string \| null;\n\tfromName?: string \| null;\n\ttoEmails?: string[];\n\tccEmails?: string[];\n\tbccEmails?: string[];\n\tmessageId?: string \| null;\n\tattachments?: string[];\n}\n\nexport interface ArchiveMetadata {\n\tformat?: string;\n\tfileCount?: number;\n\tfileList?: string[];\n\ttotalSize?: number;\n\tcompressedSize?: number \| null;\n}\n\nexport interface ImageMetadata {\n\twidth?: number;\n\theight?: number;\n\tformat?: string;\n\texif?: Record<string, string>;\n}\n\nexport interface XmlMetadata {\n\telementCount?: number;\n\tuniqueElements?: string[];\n}\n\nexport interface TextMetadata {\n\tlineCount?: number;\n\twordCount?: number;\n\tcharacterCount?: number;\n\theaders?: string[] \| null;\n\tlinks?: [string, string][] \| null;\n\tcodeBlocks?: [string, string][] \| null;\n}\n\nexport interface HtmlMetadata {\n\ttitle?: string \| null;\n\tdescription?: string \| null;\n\tkeywords?: string \| null;\n\tauthor?: string \| null;\n\tcanonical?: string \| null;\n\tbaseHref?: string \| null;\n\togTitle?: string \| null;\n\togDescription?: string \| null;\n\togImage?: string \| null;\n\togUrl?: string \| null;\n\togType?: string \| null;\n\togSiteName?: string \| null;\n\ttwitterCard?: string \| null;\n\ttwitterTitle?: string \| null;\n\ttwitterDescription?: string \| null;\n\ttwitterImage?: string \| null;\n\ttwitterSite?: string \| null;\n\ttwitterCreator?: string \| null;\n\tlinkAuthor?: string \| null;\n\tlinkLicense?: string \| null;\n\tlinkAlternate?: string \| null;\n}\n\nexport interface PdfMetadata {\n\ttitle?: string \| null;\n\tauthor?: string \| null;\n\tsubject?: string \| null;\n\tkeywords?: string \| null;\n\tcreator?: string \| null;\n\tproducer?: string \| null;\n\tcreationDate?: string \| null;\n\tmodificationDate?: string \| null;\n\tpageCount?: number;\n}\n\nexport interface PptxMetadata {\n\ttitle?: string \| null;\n\tauthor?: string \| null;\n\tdescription?: string \| null;\n\tsummary?: string \| null;\n\tfonts?: string[];\n}\n\nexport interface OcrMetadata {\n\tlanguage?: string;\n\tpsm?: number;\n\toutputFormat?: string;\n\ttableCount?: number;\n\ttableRows?: number \| null;\n\ttableCols?: number \| null;\n}\n\nexport interface ImagePreprocessingMetadata {\n\toriginalDimensions?: [number, number];\n\toriginalDpi?: [number, number];\n\ttargetDpi?: number;\n\tscaleFactor?: number;\n\tautoAdjusted?: boolean;\n\tfinalDpi?: number;\n\tnewDimensions?: [number, number] \| null;\n\tresampleMethod?: string;\n\tdimensionClamped?: boolean;\n\tcalculatedDpi?: number \| null;\n\tskippedResize?: boolean;\n\tresizeError?: string \| null;\n}\n\nexport interface ErrorMetadata {\n\terrorType?: string;\n\tmessage?: string;\n}\n\n/\n Page boundary information for chunk metadata.\n \n Tracks where a specific page's content starts and ends in the main content string,\n * enabling mapping from byte positions to page numbers. All offsets are guaranteed to be\n * at valid UTF-8 character boundaries.\n /\nexport interface PageBoundary {\n\t/* Byte offset where this page starts in the content string (UTF-8 valid boundary, inclusive) /\n\tbyteStart: number;\n\t/* Byte offset where this page ends in the content string (UTF-8 valid boundary, exclusive) /\n\tbyteEnd: number;\n\t/* Page number (1-indexed) /\n\tpageNumber: number;\n}\n\n/\n Type of paginated unit in a document.\n \n Distinguishes between different types of \"pages\":\n * - \"page\": Standard document pages (PDF, DOCX, images)\n * - \"slide\": Presentation slides (PPTX, ODP)\n * - \"sheet\": Spreadsheet sheets (XLSX, ODS)\n /\nexport type PageUnitType = \"page\" \| \"slide\" \| \"sheet\";\n\n/\n Detailed per-page metadata.\n \n Captures information about a single page/slide/sheet including dimensions,\n * content counts, and visibility state.\n /\nexport interface PageInfo {\n\t/* Page number (1-indexed) /\n\tnumber: number;\n\t/* Page title (usually for presentations) /\n\ttitle?: string \| null;\n\t/* Dimensions in points (PDF) or pixels (images): [width, height] /\n\tdimensions?: [number, number] \| null;\n\t/* Number of images on this page /\n\timageCount?: number \| null;\n\t/* Number of tables on this page /\n\ttableCount?: number \| null;\n\t/* Whether this page is hidden (e.g., in presentations) /\n\thidden?: boolean \| null;\n}\n\n/\n Page structure metadata.\n \n Contains information about pages/slides/sheets in a document, including\n * boundaries for mapping chunks to pages and detailed per-page metadata.\n /\nexport interface PageStructure {\n\t/* Total number of pages/slides/sheets /\n\ttotalCount: number;\n\t/* Type of paginated unit (page, slide, or sheet) /\n\tunitType: PageUnitType;\n\t/* Byte offset boundaries for each page /\n\tboundaries?: PageBoundary[] \| null;\n\t/* Detailed per-page metadata (optional, only when needed) /\n\tpages?: PageInfo[] \| null;\n}\n\n/\n Metadata about a chunk's position and properties in the document.\n \n Tracks where a chunk appears in the original document, including byte offsets\n * and page ranges when page tracking is enabled.\n /\nexport interface ChunkMetadata {\n\t/* Byte offset where this chunk starts in the original text (UTF-8 valid boundary) /\n\tbyteStart: number;\n\t/* Byte offset where this chunk ends in the original text (UTF-8 valid boundary) /\n\tbyteEnd: number;\n\t/* Number of tokens in this chunk (if available from embedding model) /\n\ttokenCount?: number \| null;\n\t/* Zero-based index of this chunk in the document /\n\tchunkIndex: number;\n\t/* Total number of chunks in the document /\n\ttotalChunks: number;\n\t/* First page number this chunk spans (1-indexed, only when page tracking enabled) /\n\tfirstPage?: number \| null;\n\t/* Last page number this chunk spans (1-indexed, only when page tracking enabled) /\n\tlastPage?: number \| null;\n}\n\n/\n Text chunk with optional embedding.\n \n Represents a segment of a document created by the chunking algorithm, useful for RAG and vector databases.\n /\nexport interface Chunk {\n\t/* Text content of this chunk /\n\tcontent: string;\n\n\t/* Vector embedding for this chunk (if embedding model was used) /\n\tembedding?: number[] \| null;\n\n\t/* Metadata about chunk position and properties in the document /\n\tmetadata: ChunkMetadata;\n}\n\n/\n Extracted image from document with optional OCR result.\n \n Contains image data and metadata about position, dimensions, and properties.\n /\nexport interface ExtractedImage {\n\t/* Raw image bytes as Uint8Array /\n\tdata: Uint8Array;\n\n\t/* Image format (e.g., 'png', 'jpeg', 'tiff') /\n\tformat: string;\n\n\t/* Sequential index of this image in the document (0-indexed) /\n\timageIndex: number;\n\n\t/* Page number where this image was found (1-indexed), null if unknown /\n\tpageNumber?: number \| null;\n\n\t/* Image width in pixels, null if unknown /\n\twidth?: number \| null;\n\n\t/* Image height in pixels, null if unknown /\n\theight?: number \| null;\n\n\t/* Color space (e.g., 'RGB', 'CMYK', 'Grayscale'), null if unknown /\n\tcolorspace?: string \| null;\n\n\t/* Bits per color component (e.g., 8 for 8-bit), null if unknown /\n\tbitsPerComponent?: number \| null;\n\n\t/* Whether this is a mask image (used internally by PDF) /\n\tisMask: boolean;\n\n\t/* Image description or caption if available /\n\tdescription?: string \| null;\n\n\t/* OCR extraction result if OCR was run on this image, null otherwise /\n\tocrResult?: ExtractionResult \| null;\n}\n\n/\n Content for a single page/slide/sheet.\n \n When page extraction is enabled, documents are split into per-page content\n * with associated tables and images mapped to each page. This allows for page-specific processing.\n /\nexport interface PageContent {\n\t/* Page number (1-indexed) starting from 1 /\n\tpageNumber: number;\n\n\t/* Text content extracted from this page /\n\tcontent: string;\n\n\t/* Tables found and extracted from this page /\n\ttables: Table[];\n\n\t/* Images found and extracted from this page /\n\timages: ExtractedImage[];\n}\n\n/\n Extraction result metadata.\n \n Uses a flattened discriminated union approach with format_type as the discriminator.\n * When format_type is set (e.g., \"archive\"), the corresponding format-specific fields\n * are available at the root level of the metadata object.\n \n This structure matches the Rust serialization with serde's tagged enum flattening.\n /\nexport interface Metadata {\n\tlanguage?: string \| null;\n\tdate?: string \| null;\n\tsubject?: string \| null;\n\n\tformat_type?: \"pdf\" \| \"excel\" \| \"email\" \| \"pptx\" \| \"archive\" \| \"image\" \| \"xml\" \| \"text\" \| \"html\" \| \"ocr\";\n\n\ttitle?: string \| null;\n\tauthor?: string \| null;\n\tkeywords?: string \| null;\n\tcreator?: string \| null;\n\tproducer?: string \| null;\n\tcreation_date?: string \| null;\n\tmodification_date?: string \| null;\n\tpage_count?: number;\n\n\tsheet_count?: number;\n\tsheet_names?: string[];\n\n\tfrom_email?: string \| null;\n\tfrom_name?: string \| null;\n\tto_emails?: string[];\n\tcc_emails?: string[];\n\tbcc_emails?: string[];\n\tmessage_id?: string \| null;\n\tattachments?: string[];\n\n\tdescription?: string \| null;\n\tsummary?: string \| null;\n\tfonts?: string[];\n\n\tformat?: string;\n\tfile_count?: number;\n\tfile_list?: string[];\n\ttotal_size?: number;\n\tcompressed_size?: number \| null;\n\n\twidth?: number;\n\theight?: number;\n\texif?: Record<string, string>;\n\n\telement_count?: number;\n\tunique_elements?: string[];\n\n\tline_count?: number;\n\tword_count?: number;\n\tcharacter_count?: number;\n\theaders?: string[] \| null;\n\tlinks?: [string, string][] \| null;\n\tcode_blocks?: [string, string][] \| null;\n\n\tcanonical?: string \| null;\n\tbase_href?: string \| null;\n\tog_title?: string \| null;\n\tog_description?: string \| null;\n\tog_image?: string \| null;\n\tog_url?: string \| null;\n\tog_type?: string \| null;\n\tog_site_name?: string \| null;\n\ttwitter_card?: string \| null;\n\ttwitter_title?: string \| null;\n\ttwitter_description?: string \| null;\n\ttwitter_image?: string \| null;\n\ttwitter_site?: string \| null;\n\ttwitter_creator?: string \| null;\n\tlink_author?: string \| null;\n\tlink_license?: string \| null;\n\tlink_alternate?: string \| null;\n\n\tpsm?: number;\n\toutput_format?: string;\n\ttable_count?: number;\n\ttable_rows?: number \| null;\n\ttable_cols?: number \| null;\n\n\timage_preprocessing?: ImagePreprocessingMetadata \| null;\n\n\tjson_schema?: Record<string, unknown> \| null;\n\n\tpage_structure?: PageStructure \| null;\n\n\terror?: ErrorMetadata \| null;\n\n\t/\n\t Additional fields may be added at runtime by postprocessors.\n\t * Use bracket notation to safely access unexpected properties.\n\t /\n\t[key: string]: unknown;\n}\n\n/\n Complete extraction result from document processing.\n \n Contains all extracted content, metadata, and optional processed data like chunks and images.\n * This is the primary return value from extraction functions.\n /\nexport interface ExtractionResult {\n\t/* Extracted text content from the document (main content) /\n\tcontent: string;\n\n\t/* MIME type of the input document (e.g., 'application/pdf', 'text/html') /\n\tmimeType: string;\n\n\t/* Document metadata including title, author, creation date, language, and format-specific fields /\n\tmetadata: Metadata;\n\n\t/* Tables extracted from the document (2D cell arrays with Markdown representation) /\n\ttables: Table[];\n\n\t/* Detected languages in the document (ISO 639-1 codes, e.g., ['en', 'de']), null if detection disabled /\n\tdetectedLanguages: string[] \| null;\n\n\t/* Document chunks for RAG/vector databases (if chunking was enabled), null otherwise /\n\tchunks: Chunk[] \| null;\n\n\t/* Images extracted from document with metadata (if image extraction was enabled), null otherwise /\n\timages: ExtractedImage[] \| null;\n\n\t/* Per-page content when page extraction is enabled, null otherwise. Each item contains page number, content, tables, and images. /\n\tpages?: PageContent[] \| null;\n}\n\n/* Post-processor execution stage in the extraction pipeline. /\nexport type ProcessingStage = \"early\" \| \"middle\" \| \"late\";\n\n/\n Protocol for custom post-processors that modify extraction results.\n \n Post-processors enrich or transform extraction results without failing the extraction.\n * If a post-processor throws an error, it's logged but extraction continues.\n * Only works with async extraction functions (`extractFile`, `extractBytes`, etc.).\n /\nexport interface PostProcessorProtocol {\n\t/\n\t Return the unique name of this postprocessor.\n\t \n\t @returns Unique processor name (case-sensitive, alphanumeric + underscores recommended)\n\t /\n\tname(): string;\n\n\t/\n\t Process and enrich an extraction result.\n\t \n\t Modify the result to add new metadata, transform content, or perform other enrichment.\n\t * If this throws an error, it's logged but extraction continues.\n\t \n\t @param result - ExtractionResult with extracted content, metadata, and tables\n\t * @returns Modified result with enriched data. Can be async or sync.\n\t /\n\tprocess(result: ExtractionResult): ExtractionResult \| Promise<ExtractionResult>;\n\n\t/\n\t Return the processing stage for this processor.\n\t \n\t Determines when this processor runs relative to others:\n\t * - \"early\": Runs first, before other processors (good for cleanup/normalization)\n\t * - \"middle\": Runs with other middle-stage processors (default)\n\t * - \"late\": Runs last, after others (good for final enrichment)\n\t \n\t @returns One of \"early\", \"middle\", or \"late\" (default: \"middle\")\n\t /\n\tprocessingStage?(): ProcessingStage;\n\n\t/\n\t Initialize the processor (e.g., load ML models, setup resources).\n\t \n\t Called once when the processor is first registered. Use for expensive operations.\n\t /\n\tinitialize?(): void \| Promise<void>;\n\n\t/\n\t Shutdown the processor and release resources.\n\t \n\t Called when the processor is unregistered. Use for cleanup (closing connections, freeing memory).\n\t /\n\tshutdown?(): void \| Promise<void>;\n}\n\n/\n Protocol for custom validators that check extraction results.\n \n Validators perform quality checks and fail the extraction if validation fails.\n * Unlike post-processors, validator errors cause the entire extraction to fail.\n * Useful for enforcing quality standards on extracted content.\n /\nexport interface ValidatorProtocol {\n\t/\n\t Return the unique name of this validator.\n\t \n\t @returns Unique validator name (case-sensitive, alphanumeric + underscores recommended)\n\t /\n\tname(): string;\n\n\t/\n\t Validate an extraction result.\n\t \n\t Throw an error if validation fails. The error message will be used as the extraction error.\n\t * If validation passes, return without throwing (return value is ignored).\n\t \n\t @param result - ExtractionResult to validate\n\t * @throws {Error} If validation fails (extraction will fail with this error)\n\t /\n\tvalidate(result: ExtractionResult): void \| Promise<void>;\n\n\t/\n\t Return the validation priority.\n\t \n\t Higher priority validators run first. Useful for running cheap validations (e.g., length checks)\n\t * before expensive ones (e.g., AI-based quality checks) to fail fast.\n\t \n\t @returns Priority value (higher = runs earlier, default: 50). Range: 0-1000.\n\t /\n\tpriority?(): number;\n\n\t/\n\t Check if this validator should run for a given result.\n\t \n\t Allows conditional validation based on MIME type, metadata, or content.\n\t * This is evaluated before validation, so expensive checks can be skipped for irrelevant documents.\n\t \n\t @param result - ExtractionResult to check\n\t * @returns true if validator should run, false to skip (default: true)\n\t /\n\tshouldValidate?(result: ExtractionResult): boolean;\n\n\t/\n\t Initialize the validator (e.g., load ML models, setup resources).\n\t \n\t Called once when the validator is first registered. Use for expensive operations.\n\t /\n\tinitialize?(): void \| Promise<void>;\n\n\t/\n\t Shutdown the validator and release resources.\n\t \n\t Called when the validator is unregistered. Use for cleanup (closing connections, freeing memory).\n\t /\n\tshutdown?(): void \| Promise<void>;\n}\n\n/\n OCR backend protocol for implementing custom OCR engines.\n \n This interface defines the contract for OCR backends that can be registered\n * with Kreuzberg's extraction pipeline.\n \n ## Implementation Requirements\n \n OCR backends must implement:\n * - `name()`: Return a unique backend identifier\n * - `supportedLanguages()`: Return list of supported ISO 639-1/2/3 language codes\n * - `processImage()`: Process image bytes and return extraction result\n \n ## Optional Methods\n \n - `initialize()`: Called when backend is registered (load models, etc.)\n * - `shutdown()`: Called when backend is unregistered (cleanup resources)\n \n @example\n * ```typescript\n * import { GutenOcrBackend } from '@kreuzberg/node/ocr/guten-ocr';\n * import { registerOcrBackend, extractFile } from '@kreuzberg/node';\n \n // Create and register the backend\n * const backend = new GutenOcrBackend();\n * await backend.initialize();\n * registerOcrBackend(backend);\n \n // Use with extraction\n * const result = await extractFile('scanned.pdf', null, {\n * ocr: { backend: 'guten-ocr', language: 'en' }\n * });\n * ```\n /\nexport interface OcrBackendProtocol {\n\t/\n\t Return the unique name of this OCR backend.\n\t \n\t This name is used in ExtractionConfig to select the backend:\n\t * ```typescript\n\t * { ocr: { backend: 'guten-ocr', language: 'en' } }\n\t * ```\n\t \n\t @returns Unique backend identifier (e.g., \"guten-ocr\", \"tesseract\")\n\t /\n\tname(): string;\n\n\t/\n\t Return list of supported language codes.\n\t \n\t Language codes should follow ISO 639-1 (2-letter) or ISO 639-2 (3-letter) standards.\n\t * Common codes: \"en\", \"eng\" (English), \"de\", \"deu\" (German), \"fr\", \"fra\" (French).\n\t \n\t @returns Array of supported language codes\n\t \n\t @example\n\t * ```typescript\n\t * supportedLanguages(): string[] {\n\t * return [\"en\", \"eng\", \"de\", \"deu\", \"fr\", \"fra\"];\n\t * }\n\t * ```\n\t /\n\tsupportedLanguages(): string[];\n\n\t/\n\t Process image bytes and extract text via OCR.\n\t \n\t This method receives raw image data and must return a result object with:\n\t * - `content`: Extracted text content\n\t * - `mime_type`: MIME type (usually \"text/plain\")\n\t * - `metadata`: Additional information (confidence, dimensions, etc.)\n\t * - `tables`: Optional array of detected tables\n\t \n\t @param imageBytes - Raw image data (Uint8Array) or Base64-encoded string (when called from Rust bindings)\n\t * @param language - Language code from supportedLanguages()\n\t * @returns Promise resolving to extraction result\n\t \n\t @example\n\t * ```typescript\n\t * async processImage(imageBytes: Uint8Array \| string, language: string): Promise<{\n\t * content: string;\n\t * mime_type: string;\n\t * metadata: Record<string, unknown>;\n\t * tables: unknown[];\n\t * }> {\n\t * const buffer = typeof imageBytes === \"string\" ? Buffer.from(imageBytes, \"base64\") : Buffer.from(imageBytes);\n\t * const text = await myOcrEngine.recognize(buffer, language);\n\t * return {\n\t * content: text,\n\t * mime_type: \"text/plain\",\n\t * metadata: { confidence: 0.95, language },\n\t * tables: []\n\t * };\n\t * }\n\t * ```\n\t /\n\tprocessImage(\n\t\timageBytes: Uint8Array \| string,\n\t\tlanguage: string,\n\t): Promise<{\n\t\tcontent: string;\n\t\tmime_type: string;\n\t\tmetadata: Record<string, unknown>;\n\t\ttables: unknown[];\n\t}>;\n\n\t/\n\t Initialize the OCR backend (optional).\n\t \n\t Called once when the backend is registered. Use this to:\n\t * - Load ML models\n\t * - Initialize libraries\n\t * - Validate dependencies\n\t \n\t @example\n\t * ```typescript\n\t * async initialize(): Promise<void> {\n\t * this.model = await loadModel('./path/to/model');\n\t * }\n\t * ```\n\t /\n\tinitialize?(): void \| Promise<void>;\n\n\t/\n\t Shutdown the OCR backend and release resources (optional).\n\t \n\t Called when the backend is unregistered. Use this to:\n\t * - Free model memory\n\t * - Close file handles\n\t * - Cleanup temporary files\n\t \n\t @example\n\t * ```typescript\n\t * async shutdown(): Promise<void> {\n\t * await this.model.dispose();\n\t * this.model = null;\n\t * }\n\t * ```\n\t */\n\tshutdown?(): void \| Promise<void>;\n}\n"],"mappings":";;;;;;;;;;;;;;AAAA;AAAA;","names":[]}
1	+ {"version":3,"sources":["../typescript/types.ts"],"sourcesContent":["/*\n Type definitions for Kreuzberg extraction results.\n \n These types mirror the strongly-typed Rust metadata structures,\n * providing type safety for TypeScript users.\n /\n\n// ============================================================================\n// ============================================================================\n\n/\n Tesseract OCR engine configuration options.\n \n @example\n * ```typescript\n * const config: TesseractConfig = {\n * psm: 6,\n * enableTableDetection: true,\n * tesseditCharWhitelist: '0123456789'\n * };\n * ```\n /\nexport interface TesseractConfig {\n\t/\n\t Page Segmentation Mode (0-13). Controls how Tesseract segments and recognizes text.\n\t * Common values: 3 (auto), 6 (single uniform block), 11 (sparse text).\n\t * Default: 3 (auto layout analysis).\n\t /\n\tpsm?: number;\n\n\t/\n\t Enable table detection during OCR processing.\n\t * When true, Tesseract attempts to preserve table structure in the output.\n\t * Default: false.\n\t /\n\tenableTableDetection?: boolean;\n\n\t/\n\t Whitelist of characters Tesseract should recognize.\n\t * Only these characters will be returned by the OCR engine.\n\t * Use empty string to allow all characters. Useful for constraining output to digits,\n\t * specific alphabets, or other character sets.\n\t * Default: null (recognize all).\n\t /\n\ttesseditCharWhitelist?: string;\n}\n\n/\n OCR (Optical Character Recognition) configuration.\n \n Controls which OCR engine to use and how it processes images.\n /\nexport interface OcrConfig {\n\t/* OCR backend name (e.g., 'tesseract', 'paddleocr', 'easyocr'). Required. /\n\tbackend: string;\n\n\t/* ISO 639-1/3 language code(s) for OCR (e.g., 'eng', 'fra', 'deu'). Default: 'eng'. /\n\tlanguage?: string;\n\n\t/* Tesseract engine-specific configuration options. Only used when backend is 'tesseract'. /\n\ttesseractConfig?: TesseractConfig;\n}\n\n/\n Document chunking configuration for splitting large documents.\n \n Breaks large documents into smaller, manageable chunks while preserving context.\n * Useful for RAG (Retrieval Augmented Generation) and vector database indexing.\n /\nexport interface ChunkingConfig {\n\t/* Maximum characters per chunk. Default: 4096. /\n\tmaxChars?: number;\n\n\t/* Maximum overlapping characters between consecutive chunks for context preservation. Default: 512. /\n\tmaxOverlap?: number;\n\n\t/\n\t Alternative to maxChars: chunk size using different unit.\n\t * Mutually exclusive with maxChars.\n\t /\n\tchunkSize?: number;\n\n\t/\n\t Alternative to maxOverlap: overlap amount using different unit.\n\t * Mutually exclusive with maxOverlap.\n\t /\n\tchunkOverlap?: number;\n\n\t/\n\t Named preset configuration (e.g., 'default', 'aggressive', 'minimal').\n\t * Uses preset values if neither maxChars nor chunkSize is specified.\n\t /\n\tpreset?: string;\n\n\t/* Embedding configuration for generating vector embeddings for each chunk. /\n\tembedding?: Record<string, unknown>;\n\n\t/* Enable or disable chunking. Default: true when chunking config is provided. /\n\tenabled?: boolean;\n}\n\n/\n Language detection configuration.\n \n Automatically detects the language(s) of extracted content.\n /\nexport interface LanguageDetectionConfig {\n\t/* Enable automatic language detection. Default: true. /\n\tenabled?: boolean;\n\n\t/* Minimum confidence score (0.0-1.0) for language detection. Default: 0.5. /\n\tminConfidence?: number;\n\n\t/* Detect multiple languages in the same document. Default: false. /\n\tdetectMultiple?: boolean;\n}\n\n/\n Token reduction configuration for optimizing token usage.\n \n Reduces the number of tokens in extracted content while preserving meaning.\n * Useful for reducing costs in LLM pipelines.\n /\nexport interface TokenReductionConfig {\n\t/* Reduction mode: 'aggressive' or 'conservative'. Default: 'conservative'. /\n\tmode?: string;\n\n\t/* Preserve tokens for semantically important words even in aggressive mode. Default: true. /\n\tpreserveImportantWords?: boolean;\n}\n\n/\n PDF-specific extraction configuration.\n \n Controls how PDF documents are processed.\n /\nexport interface PdfConfig {\n\t/* Extract images from PDF pages. Default: true. /\n\textractImages?: boolean;\n\n\t/* List of passwords to try for password-protected PDFs. /\n\tpasswords?: string[];\n\n\t/* Extract document metadata (title, author, creation date, etc.). Default: true. /\n\textractMetadata?: boolean;\n}\n\n/\n Image extraction and processing configuration.\n \n Controls how images are extracted and optimized from documents.\n /\nexport interface ImageExtractionConfig {\n\t/* Enable image extraction from documents. Default: true. /\n\textractImages?: boolean;\n\n\t/* Target DPI (dots per inch) for extracted images. Higher DPI = better quality but larger files. Default: 150. /\n\ttargetDpi?: number;\n\n\t/* Maximum image dimension (width or height) in pixels. Images larger than this are downscaled. Default: 2000. /\n\tmaxImageDimension?: number;\n\n\t/* Automatically adjust DPI based on image content and quality. Default: true. /\n\tautoAdjustDpi?: boolean;\n\n\t/* Minimum DPI to maintain for image quality. Default: 72. /\n\tminDpi?: number;\n\n\t/* Maximum DPI to avoid excessive file sizes. Default: 300. /\n\tmaxDpi?: number;\n}\n\n/\n Post-processor configuration for modifying extracted content.\n \n Post-processors allow customization and cleanup of extraction results\n * without failing the extraction if they encounter errors.\n /\nexport interface PostProcessorConfig {\n\t/* Enable or disable post-processing entirely. Default: true. /\n\tenabled?: boolean;\n\n\t/* List of processor names to enable (allowlist). When set, only these are used. /\n\tenabledProcessors?: string[];\n\n\t/* List of processor names to disable (denylist). These are skipped. /\n\tdisabledProcessors?: string[];\n}\n\n/\n HTML preprocessing options.\n \n Cleans HTML content before conversion to Markdown.\n /\nexport interface HtmlPreprocessingOptions {\n\t/* Enable HTML preprocessing. Default: true. /\n\tenabled?: boolean;\n\n\t/* Preset cleanup level: 'minimal' (light), 'standard' (balanced), 'aggressive' (heavy). Default: 'standard'. /\n\tpreset?: \"minimal\" \| \"standard\" \| \"aggressive\";\n\n\t/* Remove navigation menus and headers. Default: true. /\n\tremoveNavigation?: boolean;\n\n\t/* Remove form elements. Default: true. /\n\tremoveForms?: boolean;\n}\n\n/\n HTML to Markdown conversion configuration options.\n \n Controls how HTML content is converted to Markdown format, including formatting,\n * escaping, and special handling for various HTML elements.\n /\nexport interface HtmlConversionOptions {\n\t/* Heading style conversion: \"atx\" (# style), \"underlined\" (underline style), or \"atx_closed\" (# style closed). Default: \"atx\". /\n\theadingStyle?: \"atx\" \| \"underlined\" \| \"atx_closed\";\n\n\t/* List indentation type: \"spaces\" or \"tabs\". Default: \"spaces\". /\n\tlistIndentType?: \"spaces\" \| \"tabs\";\n\n\t/* Number of spaces/tabs per list indent level. Default: 4. /\n\tlistIndentWidth?: number;\n\n\t/* Bullet characters for unordered lists (e.g., '', '-', '+'). Default: ''. /\n\tbullets?: string;\n\n\t/* Markdown symbol for strong/bold emphasis: '' or '__'. Default: ''. /\n\tstrongEmSymbol?: string;\n\n\t/* Escape asterisks () in text to prevent accidental formatting. Default: false. /\n\tescapeAsterisks?: boolean;\n\n\t/** Escape underscores (_) in text to prevent accidental formatting. Default: false. /\n\tescapeUnderscores?: boolean;\n\n\t/* Escape miscellaneous special characters. Default: false. /\n\tescapeMisc?: boolean;\n\n\t/* Escape ASCII control characters. Default: false. /\n\tescapeAscii?: boolean;\n\n\t/* Default code language for syntax highlighting in code blocks (e.g., 'javascript'). Default: null. /\n\tcodeLanguage?: string;\n\n\t/* Convert HTML links to Markdown autolinks format ([text](url)). Default: true. /\n\tautolinks?: boolean;\n\n\t/* Use the HTML title element as default for links when no text is available. Default: false. /\n\tdefaultTitle?: boolean;\n\n\t/* Insert <br> tags in Markdown tables. Default: false. /\n\tbrInTables?: boolean;\n\n\t/* Use HOCR spatial table format for better table structure preservation. Default: false. /\n\thocrSpatialTables?: boolean;\n\n\t/* Highlight style for marked/highlighted text: \"double_equal\" (==text==), \"html\" (<mark>), \"bold\" (text), or \"none\". Default: \"none\". /\n\thighlightStyle?: \"double_equal\" \| \"html\" \| \"bold\" \| \"none\";\n\n\t/* Extract metadata from HTML (title, meta tags, etc.). Default: false. /\n\textractMetadata?: boolean;\n\n\t/* Whitespace handling: \"normalized\" (collapse whitespace) or \"strict\" (preserve all whitespace). Default: \"normalized\". /\n\twhitespaceMode?: \"normalized\" \| \"strict\";\n\n\t/* Remove newlines from output (convert to single line). Default: false. /\n\tstripNewlines?: boolean;\n\n\t/* Enable line wrapping at specified width. Default: true. /\n\twrap?: boolean;\n\n\t/* Maximum line width when wrapping is enabled. Default: 80. /\n\twrapWidth?: number;\n\n\t/* Convert as inline Markdown instead of block elements. Default: false. /\n\tconvertAsInline?: boolean;\n\n\t/* Markdown symbol for subscript text (e.g., '~' for ~text~). Default: '~'. /\n\tsubSymbol?: string;\n\n\t/* Markdown symbol for superscript text (e.g., '^' for ^text^). Default: '^'. /\n\tsupSymbol?: string;\n\n\t/* Newline style in output: \"spaces\" (two spaces + newline) or \"backslash\" (backslash + newline). Default: \"spaces\". /\n\tnewlineStyle?: \"spaces\" \| \"backslash\";\n\n\t/* Code block style: \"indented\" (4-space indent), \"backticks\" (```), or \"tildes\" (~~~). Default: \"backticks\". /\n\tcodeBlockStyle?: \"indented\" \| \"backticks\" \| \"tildes\";\n\n\t/* List of HTML tag names to keep as inline images (don't convert). Default: []. /\n\tkeepInlineImagesIn?: string[];\n\n\t/* Character encoding for output (e.g., 'utf-8', 'ascii'). Default: 'utf-8'. /\n\tencoding?: string;\n\n\t/* Enable debug mode for detailed conversion logging. Default: false. /\n\tdebug?: boolean;\n\n\t/* List of HTML tag names to remove entirely from output. Default: []. /\n\tstripTags?: string[];\n\n\t/* List of HTML tag names to preserve in output (don't convert to Markdown). Default: []. /\n\tpreserveTags?: string[];\n\n\t/* HTML preprocessing options for cleaning HTML before conversion. /\n\tpreprocessing?: HtmlPreprocessingOptions;\n}\n\n/* Keyword extraction algorithm type. /\nexport type KeywordAlgorithm = \"yake\" \| \"rake\";\n\n/\n YAKE (Yet Another Keyword Extractor) algorithm configuration.\n \n YAKE is an unsupervised keyword extraction method that doesn't require training data.\n /\nexport interface YakeParams {\n\t/* Window size for co-occurrence analysis (number of words to consider). Default: 3. /\n\twindowSize?: number;\n}\n\n/\n RAKE (Rapid Automatic Keyword Extraction) algorithm configuration.\n \n RAKE extracts keywords based on word co-occurrence and statistical measures.\n /\nexport interface RakeParams {\n\t/* Minimum word length to consider as keyword. Default: 3. /\n\tminWordLength?: number;\n\n\t/* Maximum number of words per keyword phrase. Default: 3. /\n\tmaxWordsPerPhrase?: number;\n}\n\n/\n Keyword extraction configuration.\n \n Extracts important keywords/phrases from document content using YAKE or RAKE algorithms.\n /\nexport interface KeywordConfig {\n\t/* Extraction algorithm: \"yake\" or \"rake\". Default: \"yake\". /\n\talgorithm?: KeywordAlgorithm;\n\n\t/* Maximum number of keywords to extract. Default: 10. /\n\tmaxKeywords?: number;\n\n\t/* Minimum relevance score (0.0-1.0) for keywords. Keywords below this are filtered out. Default: 0.1. /\n\tminScore?: number;\n\n\t/* N-gram range: [min_length, max_length] for phrase keywords (e.g., [1, 3] for 1-3 word phrases). Default: [1, 3]. /\n\tngramRange?: [number, number];\n\n\t/* Language for keyword extraction (e.g., 'en', 'de', 'fr'). Default: 'en'. /\n\tlanguage?: string;\n\n\t/* YAKE algorithm-specific parameters. Only used when algorithm is \"yake\". /\n\tyakeParams?: YakeParams;\n\n\t/* RAKE algorithm-specific parameters. Only used when algorithm is \"rake\". /\n\trakeParams?: RakeParams;\n}\n\n/\n Page tracking and extraction configuration.\n \n Controls how pages/slides/sheets are extracted and tracked in the document.\n * Page range information in chunk metadata (first_page/last_page) is automatically\n * enabled when page boundaries are available and chunking is configured.\n /\nexport interface PageConfig {\n\t/* Extract pages as separate array (ExtractionResult.pages) /\n\textractPages?: boolean;\n\t/* Insert page markers in main content string /\n\tinsertPageMarkers?: boolean;\n\t/* Page marker format (use {page_num} placeholder) /\n\tmarkerFormat?: string;\n}\n\n/\n Main extraction configuration interface.\n \n Combines all sub-configurations for document extraction, OCR, chunking, post-processing, etc.\n * All fields are optional and use sensible defaults.\n /\nexport interface ExtractionConfig {\n\t/* Enable caching of extraction results for identical inputs. Default: true. /\n\tuseCache?: boolean;\n\n\t/* Enable quality processing filters to improve extraction reliability. Default: false. /\n\tenableQualityProcessing?: boolean;\n\n\t/* OCR configuration for text extraction from images. Only used when document contains images or forceOcr is true. /\n\tocr?: OcrConfig;\n\n\t/* Force OCR processing even for documents with selectable text. Useful for scanned documents. Default: false. /\n\tforceOcr?: boolean;\n\n\t/* Chunking configuration for splitting documents into smaller pieces for RAG or vector DB. /\n\tchunking?: ChunkingConfig;\n\n\t/* Image extraction and optimization configuration. /\n\timages?: ImageExtractionConfig;\n\n\t/* PDF-specific extraction options (passwords, metadata, etc.). /\n\tpdfOptions?: PdfConfig;\n\n\t/* Token reduction configuration for optimizing token usage in LLM pipelines. /\n\ttokenReduction?: TokenReductionConfig;\n\n\t/* Language detection configuration for automatic language identification. /\n\tlanguageDetection?: LanguageDetectionConfig;\n\n\t/* Post-processor configuration for customizing extraction results. /\n\tpostprocessor?: PostProcessorConfig;\n\n\t/* HTML to Markdown conversion options for HTML content. /\n\thtmlOptions?: HtmlConversionOptions;\n\n\t/* Keyword extraction configuration for extracting important phrases. /\n\tkeywords?: KeywordConfig;\n\n\t/* Page tracking and extraction configuration for multi-page documents. /\n\tpages?: PageConfig;\n\n\t/* Maximum number of concurrent extractions in batch operations. Default: 4. /\n\tmaxConcurrentExtractions?: number;\n}\n\n/\n Extracted table data from document.\n \n Contains both cell data and Markdown representation for easy display and processing.\n /\nexport interface Table {\n\t/* 2D array of cell contents (rows × columns) /\n\tcells: string[][];\n\n\t/* Markdown representation of the table for display or parsing /\n\tmarkdown: string;\n\n\t/* Page number where this table was found (1-indexed) /\n\tpageNumber: number;\n}\n\nexport interface ExcelMetadata {\n\tsheetCount?: number;\n\tsheetNames?: string[];\n}\n\nexport interface EmailMetadata {\n\tfromEmail?: string \| null;\n\tfromName?: string \| null;\n\ttoEmails?: string[];\n\tccEmails?: string[];\n\tbccEmails?: string[];\n\tmessageId?: string \| null;\n\tattachments?: string[];\n}\n\nexport interface ArchiveMetadata {\n\tformat?: string;\n\tfileCount?: number;\n\tfileList?: string[];\n\ttotalSize?: number;\n\tcompressedSize?: number \| null;\n}\n\nexport interface ImageMetadata {\n\twidth?: number;\n\theight?: number;\n\tformat?: string;\n\texif?: Record<string, string>;\n}\n\nexport interface XmlMetadata {\n\telementCount?: number;\n\tuniqueElements?: string[];\n}\n\nexport interface TextMetadata {\n\tlineCount?: number;\n\twordCount?: number;\n\tcharacterCount?: number;\n\theaders?: string[] \| null;\n\tlinks?: [string, string][] \| null;\n\tcodeBlocks?: [string, string][] \| null;\n}\n\nexport interface HtmlMetadata {\n\ttitle?: string \| null;\n\tdescription?: string \| null;\n\tkeywords?: string \| null;\n\tauthor?: string \| null;\n\tcanonical?: string \| null;\n\tbaseHref?: string \| null;\n\togTitle?: string \| null;\n\togDescription?: string \| null;\n\togImage?: string \| null;\n\togUrl?: string \| null;\n\togType?: string \| null;\n\togSiteName?: string \| null;\n\ttwitterCard?: string \| null;\n\ttwitterTitle?: string \| null;\n\ttwitterDescription?: string \| null;\n\ttwitterImage?: string \| null;\n\ttwitterSite?: string \| null;\n\ttwitterCreator?: string \| null;\n\tlinkAuthor?: string \| null;\n\tlinkLicense?: string \| null;\n\tlinkAlternate?: string \| null;\n}\n\nexport interface PdfMetadata {\n\ttitle?: string \| null;\n\tauthor?: string \| null;\n\tsubject?: string \| null;\n\tkeywords?: string \| null;\n\tcreator?: string \| null;\n\tproducer?: string \| null;\n\tcreationDate?: string \| null;\n\tmodificationDate?: string \| null;\n\tpageCount?: number;\n}\n\nexport interface PptxMetadata {\n\ttitle?: string \| null;\n\tauthor?: string \| null;\n\tdescription?: string \| null;\n\tsummary?: string \| null;\n\tfonts?: string[];\n}\n\nexport interface OcrMetadata {\n\tlanguage?: string;\n\tpsm?: number;\n\toutputFormat?: string;\n\ttableCount?: number;\n\ttableRows?: number \| null;\n\ttableCols?: number \| null;\n}\n\nexport interface ImagePreprocessingMetadata {\n\toriginalDimensions?: [number, number];\n\toriginalDpi?: [number, number];\n\ttargetDpi?: number;\n\tscaleFactor?: number;\n\tautoAdjusted?: boolean;\n\tfinalDpi?: number;\n\tnewDimensions?: [number, number] \| null;\n\tresampleMethod?: string;\n\tdimensionClamped?: boolean;\n\tcalculatedDpi?: number \| null;\n\tskippedResize?: boolean;\n\tresizeError?: string \| null;\n}\n\nexport interface ErrorMetadata {\n\terrorType?: string;\n\tmessage?: string;\n}\n\n/\n Page boundary information for chunk metadata.\n \n Tracks where a specific page's content starts and ends in the main content string,\n * enabling mapping from byte positions to page numbers. All offsets are guaranteed to be\n * at valid UTF-8 character boundaries.\n /\nexport interface PageBoundary {\n\t/* Byte offset where this page starts in the content string (UTF-8 valid boundary, inclusive) /\n\tbyteStart: number;\n\t/* Byte offset where this page ends in the content string (UTF-8 valid boundary, exclusive) /\n\tbyteEnd: number;\n\t/* Page number (1-indexed) /\n\tpageNumber: number;\n}\n\n/\n Type of paginated unit in a document.\n \n Distinguishes between different types of \"pages\":\n * - \"page\": Standard document pages (PDF, DOCX, images)\n * - \"slide\": Presentation slides (PPTX, ODP)\n * - \"sheet\": Spreadsheet sheets (XLSX, ODS)\n /\nexport type PageUnitType = \"page\" \| \"slide\" \| \"sheet\";\n\n/\n Detailed per-page metadata.\n \n Captures information about a single page/slide/sheet including dimensions,\n * content counts, and visibility state.\n /\nexport interface PageInfo {\n\t/* Page number (1-indexed) /\n\tnumber: number;\n\t/* Page title (usually for presentations) /\n\ttitle?: string \| null;\n\t/* Dimensions in points (PDF) or pixels (images): [width, height] /\n\tdimensions?: [number, number] \| null;\n\t/* Number of images on this page /\n\timageCount?: number \| null;\n\t/* Number of tables on this page /\n\ttableCount?: number \| null;\n\t/* Whether this page is hidden (e.g., in presentations) /\n\thidden?: boolean \| null;\n}\n\n/\n Page structure metadata.\n \n Contains information about pages/slides/sheets in a document, including\n * boundaries for mapping chunks to pages and detailed per-page metadata.\n /\nexport interface PageStructure {\n\t/* Total number of pages/slides/sheets /\n\ttotalCount: number;\n\t/* Type of paginated unit (page, slide, or sheet) /\n\tunitType: PageUnitType;\n\t/* Byte offset boundaries for each page /\n\tboundaries?: PageBoundary[] \| null;\n\t/* Detailed per-page metadata (optional, only when needed) /\n\tpages?: PageInfo[] \| null;\n}\n\n/\n Metadata about a chunk's position and properties in the document.\n \n Tracks where a chunk appears in the original document, including byte offsets\n * and page ranges when page tracking is enabled.\n /\nexport interface ChunkMetadata {\n\t/* Byte offset where this chunk starts in the original text (UTF-8 valid boundary) /\n\tbyteStart: number;\n\t/* Byte offset where this chunk ends in the original text (UTF-8 valid boundary) /\n\tbyteEnd: number;\n\t/* Number of tokens in this chunk (if available from embedding model) /\n\ttokenCount?: number \| null;\n\t/* Zero-based index of this chunk in the document /\n\tchunkIndex: number;\n\t/* Total number of chunks in the document /\n\ttotalChunks: number;\n\t/* First page number this chunk spans (1-indexed, only when page tracking enabled) /\n\tfirstPage?: number \| null;\n\t/* Last page number this chunk spans (1-indexed, only when page tracking enabled) /\n\tlastPage?: number \| null;\n}\n\n/\n Text chunk with optional embedding.\n \n Represents a segment of a document created by the chunking algorithm, useful for RAG and vector databases.\n /\nexport interface Chunk {\n\t/* Text content of this chunk /\n\tcontent: string;\n\n\t/* Vector embedding for this chunk (if embedding model was used) /\n\tembedding?: number[] \| null;\n\n\t/* Metadata about chunk position and properties in the document /\n\tmetadata: ChunkMetadata;\n}\n\n/\n Extracted image from document with optional OCR result.\n \n Contains image data and metadata about position, dimensions, and properties.\n /\nexport interface ExtractedImage {\n\t/* Raw image bytes as Uint8Array /\n\tdata: Uint8Array;\n\n\t/* Image format (e.g., 'png', 'jpeg', 'tiff') /\n\tformat: string;\n\n\t/* Sequential index of this image in the document (0-indexed) /\n\timageIndex: number;\n\n\t/* Page number where this image was found (1-indexed), null if unknown /\n\tpageNumber?: number \| null;\n\n\t/* Image width in pixels, null if unknown /\n\twidth?: number \| null;\n\n\t/* Image height in pixels, null if unknown /\n\theight?: number \| null;\n\n\t/* Color space (e.g., 'RGB', 'CMYK', 'Grayscale'), null if unknown /\n\tcolorspace?: string \| null;\n\n\t/* Bits per color component (e.g., 8 for 8-bit), null if unknown /\n\tbitsPerComponent?: number \| null;\n\n\t/* Whether this is a mask image (used internally by PDF) /\n\tisMask: boolean;\n\n\t/* Image description or caption if available /\n\tdescription?: string \| null;\n\n\t/* OCR extraction result if OCR was run on this image, null otherwise /\n\tocrResult?: ExtractionResult \| null;\n}\n\n/\n Content for a single page/slide/sheet.\n \n When page extraction is enabled, documents are split into per-page content\n * with associated tables and images mapped to each page. This allows for page-specific processing.\n /\nexport interface PageContent {\n\t/* Page number (1-indexed) starting from 1 /\n\tpageNumber: number;\n\n\t/* Text content extracted from this page /\n\tcontent: string;\n\n\t/* Tables found and extracted from this page /\n\ttables: Table[];\n\n\t/* Images found and extracted from this page /\n\timages: ExtractedImage[];\n}\n\n/\n Extraction result metadata.\n \n Uses a flattened discriminated union approach with format_type as the discriminator.\n * When format_type is set (e.g., \"archive\"), the corresponding format-specific fields\n * are available at the root level of the metadata object.\n \n This structure matches the Rust serialization with serde's tagged enum flattening.\n /\nexport interface Metadata {\n\tlanguage?: string \| null;\n\tdate?: string \| null;\n\tsubject?: string \| null;\n\n\tformat_type?: \"pdf\" \| \"excel\" \| \"email\" \| \"pptx\" \| \"archive\" \| \"image\" \| \"xml\" \| \"text\" \| \"html\" \| \"ocr\";\n\n\ttitle?: string \| null;\n\tauthor?: string \| null;\n\tkeywords?: string \| null;\n\tcreator?: string \| null;\n\tproducer?: string \| null;\n\tcreation_date?: string \| null;\n\tmodification_date?: string \| null;\n\tpage_count?: number;\n\n\tsheet_count?: number;\n\tsheet_names?: string[];\n\n\tfrom_email?: string \| null;\n\tfrom_name?: string \| null;\n\tto_emails?: string[];\n\tcc_emails?: string[];\n\tbcc_emails?: string[];\n\tmessage_id?: string \| null;\n\tattachments?: string[];\n\n\tdescription?: string \| null;\n\tsummary?: string \| null;\n\tfonts?: string[];\n\n\tformat?: string;\n\tfile_count?: number;\n\tfile_list?: string[];\n\ttotal_size?: number;\n\tcompressed_size?: number \| null;\n\n\twidth?: number;\n\theight?: number;\n\texif?: Record<string, string>;\n\n\telement_count?: number;\n\tunique_elements?: string[];\n\n\tline_count?: number;\n\tword_count?: number;\n\tcharacter_count?: number;\n\theaders?: string[] \| null;\n\tlinks?: [string, string][] \| null;\n\tcode_blocks?: [string, string][] \| null;\n\n\tcanonical?: string \| null;\n\tbase_href?: string \| null;\n\tog_title?: string \| null;\n\tog_description?: string \| null;\n\tog_image?: string \| null;\n\tog_url?: string \| null;\n\tog_type?: string \| null;\n\tog_site_name?: string \| null;\n\ttwitter_card?: string \| null;\n\ttwitter_title?: string \| null;\n\ttwitter_description?: string \| null;\n\ttwitter_image?: string \| null;\n\ttwitter_site?: string \| null;\n\ttwitter_creator?: string \| null;\n\tlink_author?: string \| null;\n\tlink_license?: string \| null;\n\tlink_alternate?: string \| null;\n\n\tpsm?: number;\n\toutput_format?: string;\n\ttable_count?: number;\n\ttable_rows?: number \| null;\n\ttable_cols?: number \| null;\n\n\timage_preprocessing?: ImagePreprocessingMetadata \| null;\n\n\tjson_schema?: Record<string, unknown> \| null;\n\n\tpage_structure?: PageStructure \| null;\n\n\terror?: ErrorMetadata \| null;\n\n\t/\n\t Additional fields may be added at runtime by postprocessors.\n\t * Use bracket notation to safely access unexpected properties.\n\t /\n\t[key: string]: unknown;\n}\n\n/\n Complete extraction result from document processing.\n \n Contains all extracted content, metadata, and optional processed data like chunks and images.\n * This is the primary return value from extraction functions.\n /\nexport interface ExtractionResult {\n\t/* Extracted text content from the document (main content) /\n\tcontent: string;\n\n\t/* MIME type of the input document (e.g., 'application/pdf', 'text/html') /\n\tmimeType: string;\n\n\t/* Document metadata including title, author, creation date, language, and format-specific fields /\n\tmetadata: Metadata;\n\n\t/* Tables extracted from the document (2D cell arrays with Markdown representation) /\n\ttables: Table[];\n\n\t/* Detected languages in the document (ISO 639-1 codes, e.g., ['en', 'de']), null if detection disabled /\n\tdetectedLanguages: string[] \| null;\n\n\t/* Document chunks for RAG/vector databases (if chunking was enabled), null otherwise /\n\tchunks: Chunk[] \| null;\n\n\t/* Images extracted from document with metadata (if image extraction was enabled), null otherwise /\n\timages: ExtractedImage[] \| null;\n\n\t/* Per-page content when page extraction is enabled, null otherwise. Each item contains page number, content, tables, and images. /\n\tpages?: PageContent[] \| null;\n}\n\n/* Post-processor execution stage in the extraction pipeline. /\nexport type ProcessingStage = \"early\" \| \"middle\" \| \"late\";\n\n/\n Protocol for custom post-processors that modify extraction results.\n \n Post-processors enrich or transform extraction results without failing the extraction.\n * If a post-processor throws an error, it's logged but extraction continues.\n * Only works with async extraction functions (`extractFile`, `extractBytes`, etc.).\n /\nexport interface PostProcessorProtocol {\n\t/\n\t Return the unique name of this postprocessor.\n\t \n\t @returns Unique processor name (case-sensitive, alphanumeric + underscores recommended)\n\t /\n\tname(): string;\n\n\t/\n\t Process and enrich an extraction result.\n\t \n\t Modify the result to add new metadata, transform content, or perform other enrichment.\n\t * If this throws an error, it's logged but extraction continues.\n\t \n\t @param result - ExtractionResult with extracted content, metadata, and tables\n\t * @returns Modified result with enriched data. Can be async or sync.\n\t /\n\tprocess(result: ExtractionResult): ExtractionResult \| Promise<ExtractionResult>;\n\n\t/\n\t Return the processing stage for this processor.\n\t \n\t Determines when this processor runs relative to others:\n\t * - \"early\": Runs first, before other processors (good for cleanup/normalization)\n\t * - \"middle\": Runs with other middle-stage processors (default)\n\t * - \"late\": Runs last, after others (good for final enrichment)\n\t \n\t @returns One of \"early\", \"middle\", or \"late\" (default: \"middle\")\n\t /\n\tprocessingStage?(): ProcessingStage;\n\n\t/\n\t Initialize the processor (e.g., load ML models, setup resources).\n\t \n\t Called once when the processor is first registered. Use for expensive operations.\n\t /\n\tinitialize?(): void \| Promise<void>;\n\n\t/\n\t Shutdown the processor and release resources.\n\t \n\t Called when the processor is unregistered. Use for cleanup (closing connections, freeing memory).\n\t /\n\tshutdown?(): void \| Promise<void>;\n}\n\n/\n Protocol for custom validators that check extraction results.\n \n Validators perform quality checks and fail the extraction if validation fails.\n * Unlike post-processors, validator errors cause the entire extraction to fail.\n * Useful for enforcing quality standards on extracted content.\n /\nexport interface ValidatorProtocol {\n\t/\n\t Return the unique name of this validator.\n\t \n\t @returns Unique validator name (case-sensitive, alphanumeric + underscores recommended)\n\t /\n\tname(): string;\n\n\t/\n\t Validate an extraction result.\n\t \n\t Throw an error if validation fails. The error message will be used as the extraction error.\n\t * If validation passes, return without throwing (return value is ignored).\n\t \n\t @param result - ExtractionResult to validate\n\t * @throws {Error} If validation fails (extraction will fail with this error)\n\t /\n\tvalidate(result: ExtractionResult): void \| Promise<void>;\n\n\t/\n\t Return the validation priority.\n\t \n\t Higher priority validators run first. Useful for running cheap validations (e.g., length checks)\n\t * before expensive ones (e.g., AI-based quality checks) to fail fast.\n\t \n\t @returns Priority value (higher = runs earlier, default: 50). Range: 0-1000.\n\t /\n\tpriority?(): number;\n\n\t/\n\t Check if this validator should run for a given result.\n\t \n\t Allows conditional validation based on MIME type, metadata, or content.\n\t * This is evaluated before validation, so expensive checks can be skipped for irrelevant documents.\n\t \n\t @param result - ExtractionResult to check\n\t * @returns true if validator should run, false to skip (default: true)\n\t /\n\tshouldValidate?(result: ExtractionResult): boolean;\n\n\t/\n\t Initialize the validator (e.g., load ML models, setup resources).\n\t \n\t Called once when the validator is first registered. Use for expensive operations.\n\t /\n\tinitialize?(): void \| Promise<void>;\n\n\t/\n\t Shutdown the validator and release resources.\n\t \n\t Called when the validator is unregistered. Use for cleanup (closing connections, freeing memory).\n\t /\n\tshutdown?(): void \| Promise<void>;\n}\n\n/\n OCR backend protocol for implementing custom OCR engines.\n \n This interface defines the contract for OCR backends that can be registered\n * with Kreuzberg's extraction pipeline.\n \n ## Implementation Requirements\n \n OCR backends must implement:\n * - `name()`: Return a unique backend identifier\n * - `supportedLanguages()`: Return list of supported ISO 639-1/2/3 language codes\n * - `processImage()`: Process image bytes and return extraction result\n \n ## Optional Methods\n \n - `initialize()`: Called when backend is registered (load models, etc.)\n * - `shutdown()`: Called when backend is unregistered (cleanup resources)\n \n @example\n * ```typescript\n * import { GutenOcrBackend } from '@kreuzberg/node/ocr/guten-ocr';\n * import { registerOcrBackend, extractFile } from '@kreuzberg/node';\n \n // Create and register the backend\n * const backend = new GutenOcrBackend();\n * await backend.initialize();\n * registerOcrBackend(backend);\n \n // Use with extraction\n * const result = await extractFile('scanned.pdf', null, {\n * ocr: { backend: 'guten-ocr', language: 'en' }\n * });\n * ```\n /\nexport interface OcrBackendProtocol {\n\t/\n\t Return the unique name of this OCR backend.\n\t \n\t This name is used in ExtractionConfig to select the backend:\n\t * ```typescript\n\t * { ocr: { backend: 'guten-ocr', language: 'en' } }\n\t * ```\n\t \n\t @returns Unique backend identifier (e.g., \"guten-ocr\", \"tesseract\")\n\t /\n\tname(): string;\n\n\t/\n\t Return list of supported language codes.\n\t \n\t Language codes should follow ISO 639-1 (2-letter) or ISO 639-2 (3-letter) standards.\n\t * Common codes: \"en\", \"eng\" (English), \"de\", \"deu\" (German), \"fr\", \"fra\" (French).\n\t \n\t @returns Array of supported language codes\n\t \n\t @example\n\t * ```typescript\n\t * supportedLanguages(): string[] {\n\t * return [\"en\", \"eng\", \"de\", \"deu\", \"fr\", \"fra\"];\n\t * }\n\t * ```\n\t /\n\tsupportedLanguages(): string[];\n\n\t/\n\t Process image bytes and extract text via OCR.\n\t \n\t This method receives raw image data and must return a result object with:\n\t * - `content`: Extracted text content\n\t * - `mime_type`: MIME type (usually \"text/plain\")\n\t * - `metadata`: Additional information (confidence, dimensions, etc.)\n\t * - `tables`: Optional array of detected tables\n\t \n\t @param imageBytes - Raw image data (Uint8Array) or Base64-encoded string (when called from Rust bindings)\n\t * @param language - Language code from supportedLanguages()\n\t * @returns Promise resolving to extraction result\n\t \n\t @example\n\t * ```typescript\n\t * async processImage(imageBytes: Uint8Array \| string, language: string): Promise<{\n\t * content: string;\n\t * mime_type: string;\n\t * metadata: Record<string, unknown>;\n\t * tables: unknown[];\n\t * }> {\n\t * const buffer = typeof imageBytes === \"string\" ? Buffer.from(imageBytes, \"base64\") : Buffer.from(imageBytes);\n\t * const text = await myOcrEngine.recognize(buffer, language);\n\t * return {\n\t * content: text,\n\t * mime_type: \"text/plain\",\n\t * metadata: { confidence: 0.95, language },\n\t * tables: []\n\t * };\n\t * }\n\t * ```\n\t /\n\tprocessImage(\n\t\timageBytes: Uint8Array \| string,\n\t\tlanguage: string,\n\t): Promise<{\n\t\tcontent: string;\n\t\tmime_type: string;\n\t\tmetadata: Record<string, unknown>;\n\t\ttables: unknown[];\n\t}>;\n\n\t/\n\t Initialize the OCR backend (optional).\n\t \n\t Called once when the backend is registered. Use this to:\n\t * - Load ML models\n\t * - Initialize libraries\n\t * - Validate dependencies\n\t \n\t @example\n\t * ```typescript\n\t * async initialize(): Promise<void> {\n\t * this.model = await loadModel('./path/to/model');\n\t * }\n\t * ```\n\t /\n\tinitialize?(): void \| Promise<void>;\n\n\t/\n\t Shutdown the OCR backend and release resources (optional).\n\t \n\t Called when the backend is unregistered. Use this to:\n\t * - Free model memory\n\t * - Close file handles\n\t * - Cleanup temporary files\n\t \n\t @example\n\t * ```typescript\n\t * async shutdown(): Promise<void> {\n\t * await this.model.dispose();\n\t * this.model = null;\n\t * }\n\t * ```\n\t /\n\tshutdown?(): void \| Promise<void>;\n}\n\n/\n Result of error message classification into error codes.\n \n Provides classification details including the error code, name,\n * description, and confidence score for the classification.\n \n @example\n * ```typescript\n * import { classifyError, ErrorCode } from '@kreuzberg/node';\n \n const result = classifyError(\"File not found in read operation\");\n * if (result.code === ErrorCode.IoError) {\n * console.error(`I/O Error: ${result.description}`);\n * console.log(`Confidence: ${result.confidence}`);\n * }\n * ```\n /\nexport interface ErrorClassification {\n\t/\n\t The numeric error code (0-7) representing the error type.\n\t /\n\tcode: number;\n\n\t/\n\t The human-readable name of the error code (e.g., \"validation\", \"ocr\").\n\t /\n\tname: string;\n\n\t/\n\t A brief description of the error type.\n\t /\n\tdescription: string;\n\n\t/\n\t Confidence score (0.0-1.0) indicating how certain the classification is.\n\t * Higher values indicate higher confidence in the classification.\n\t */\n\tconfidence: number;\n}\n"],"mappings":";;;;;;;;;;;;;;AAAA;AAAA;","names":[]}