/**
 * Smart text chunker for logosdb-mcp-server.
 *
 * Three strategies, selected by `mode` (or auto-detected from `filePath`):
 *
 *   "auto"    — (default) picks the best strategy per file extension:
 *                 · code files (.ts .py .go .rs .java .c .cpp …) → "line"
 *                 · docs (.md .rst)                               → "section"
 *                 · config/data (.json .yaml .toml .txt …)        → "legacy"
 *
 *   "line"    — sliding line-window (code files).
 *               Target ~50 lines per chunk, ~10-line overlap.
 *               Prefers blank-line boundaries at window edges to avoid
 *               splitting mid-function. Best for retrieval precision on code.
 *
 *   "section" — heading-aware (Markdown / RST).
 *               Splits on ATX headings (# … ###### …) and Setext underlines
 *               (=== / ---). Small consecutive sections are merged up to
 *               `targetChars`. Large single sections are sub-split with the
 *               paragraph chunker. Best for docs and README-style files.
 *
 *   "legacy"  — original paragraph / character merge (any file type).
 *               Splits on blank lines, merges short paragraphs up to
 *               `targetChars` (default 800), carries `overlapChars` (default
 *               100) into the next chunk. Long single paragraphs are sub-split
 *               into character-window slices. Use this mode when you want the
 *               old behaviour for A/B testing, compatibility with an existing
 *               namespace, or for dense config/data files where paragraph merge
 *               is appropriate. Enable globally with `LOGOSDB_CHUNK_MODE=legacy`
 *               or per-call with `chunking: "legacy"` on `logosdb_index_file`.
 *
 * Related: GitHub issue #98.
 */
export interface Chunk {
    text: string;
    /** approximate character offset in original text */
    offset: number;
}
export type ChunkMode = 'auto' | 'line' | 'section' | 'legacy';
export interface ChunkOptions {
    /** Target character count per chunk (legacy / section modes). Default: 800. */
    targetChars?: number;
    /** Character overlap carried into the next chunk (legacy mode). Default: 100. */
    overlapChars?: number;
    /** Chunking strategy. Default: "auto". */
    mode?: ChunkMode;
    /**
     * Absolute or relative file path — used by "auto" to pick the right strategy.
     * Ignored unless mode is "auto".
     */
    filePath?: string;
    /** Lines per chunk (line mode). Default: 50. */
    linesPerChunk?: number;
    /** Lines of overlap between consecutive line-window chunks. Default: 10. */
    lineOverlap?: number;
}
/**
 * Chunk `text` according to `opts`.
 * The old two-argument signature `chunk(text, targetChars, overlapChars)` is
 * preserved for callers that have not yet migrated.
 */
export declare function chunk(text: string, opts?: ChunkOptions | number, _legacyOverlap?: number): Chunk[];
//# sourceMappingURL=chunker.d.ts.map