mongodocs-mcp/dist/core/smart-chunker.d.ts

Lightning-fast semantic search for MongoDB documentation via Model Context Protocol. 10,000+ documents, <500ms search.

/**
 * Smart Document Chunker - Adaptive chunking for any content type
 *
 * Merges the best of:
 * - DocumentChunker (basic chunking)
 * - EnhancedDocumentChunker (quality scoring)
 *
 * Features:
 * - Automatic content type detection
 * - Optimal chunk sizing for voyage-3 vs voyage-code-3
 * - Semantic boundary preservation
 * - Code block integrity
 */
import { Document, ChunkedDocument } from '../types/index.js';
export interface ChunkOptions {
    targetSize?: number;
    maxSize?: number;
    minSize?: number;
    preserveCode?: boolean;
    preserveExamples?: boolean;
    semanticBoundaries?: boolean;
    overlap?: number;
    chunkingStrategy?: 'fixed' | 'semantic' | 'proposition';
    semanticThreshold?: number;
    breakpointType?: 'percentile' | 'standard_deviation' | 'interquartile' | 'gradient';
}
export declare class SmartChunker {
    private encoder;
    private qualityScorer;
    private readonly DEFAULT_OPTIONS;
    /**
     * Chunk multiple documents intelligently
     */
    chunkDocuments(documents: Document[], options?: ChunkOptions): Promise<ChunkedDocument[]>;
    /**
     * Adapt chunking options based on content type
     */
    private adaptOptionsForContent;
    /**
     * Chunk a single document with quality awareness
     */
    private chunkDocument;
    /**
     * Parse document into logical sections
     */
    private parseDocumentStructure;
    /**
     * Chunk a section intelligently
     */
    private chunkSection;
    /**
     * Split content by sentences
     */
    private splitBySentences;
    /**
     * Extract code blocks from content
     */
    private extractCodeBlocks;
    /**
     * Split large code block carefully
     */
    private splitLargeCodeBlock;
    /**
     * Get overlap text from previous chunk
     */
    private getOverlapText;
    /**
     * Create a chunk with metadata
     */
    private createChunk;
    /**
     * Detect if content contains code
     */
    private detectCode;
    /**
     * Detect header in markdown or RST
     */
    private detectHeader;
    /**
     * Split text into sentences
     */
    private splitIntoSentences;
    /**
     * Generate content hash
     */
    hashContent(content: string): string;
}
//# sourceMappingURL=smart-chunker.d.ts.map