/**
 * ChunkStrategy — Document chunking strategies.
 *
 * Provides multiple strategies for splitting content into chunks:
 * 1. By heading (default) — each heading section is a chunk
 * 2. By token budget — split into chunks of maxTokens size
 * 3. With overlap window — chunks overlap by overlapTokens
 *
 * Token estimation: ~4 chars per token (rough approximation for mixed CJK/Latin).
 */
import type { ParsedSection } from './document-parser.js';
export interface ChunkOptions {
    /** Maximum tokens per chunk (default: 512) */
    maxTokens?: number;
    /** Overlap tokens between consecutive chunks (default: 0) */
    overlapTokens?: number;
}
export interface Chunk {
    content: string;
    index: number;
    tokenEstimate: number;
    metadata: Record<string, unknown>;
}
/** Approximate token count: ~4 chars per token for mixed content */
declare function estimateTokens(text: string): number;
export declare class ChunkStrategy {
    private maxTokens;
    private overlapTokens;
    constructor(options?: ChunkOptions);
    /**
     * Chunk by heading — each ParsedSection becomes one chunk.
     * This is the default strategy.
     */
    chunkByHeading(sections: ParsedSection[]): Chunk[];
    /**
     * Chunk by token budget — split text into chunks respecting maxTokens.
     * Splits at paragraph boundaries when possible.
     */
    chunkByTokenBudget(text: string): Chunk[];
    /**
     * Chunk sections by token budget — applies token budget to each section,
     * splitting large sections into multiple chunks.
     */
    chunkSectionsByTokenBudget(sections: ParsedSection[]): Chunk[];
    private createChunk;
    private splitLargeParagraph;
    /**
     * Get overlap text from the tail of content.
     * Returns approximately overlapTokens worth of trailing text.
     */
    private getOverlapText;
}
export { estimateTokens };
//# sourceMappingURL=chunk-strategy.d.ts.map