/**
 * Text extraction for the two document formats a CV or a receipt comes in
 * (issue #263). A 5 MiB PDF costs ~6.6 MB of base64 in the model's context for
 * a content that fits in a few KB of text; extracting server-side is what
 * makes reading resumes in batches affordable.
 *
 * - PDF: `unpdf` (a serverless build of pdf.js, no native binary, ~2 MB).
 * - DOCX: the file is a zip whose `word/document.xml` holds the text; the zip
 *   is read here with `node:zlib` only (STORED and DEFLATE entries) — no
 *   dependency for a 60-line format.
 */
export interface ExtractedText {
    text: string;
    /** Page count for PDFs; absent for DOCX. */
    pages?: number;
}
export declare const IMAGE_MIMES: Set<string>;
export declare function isImageMime(mime: string): boolean;
export declare function isPdfMime(mime: string): boolean;
/**
 * Is this a DOCX? BoondManager serves `.docx` resumes as `application/msword`
 * (observed live on 2026-09-26, issue #311) — the legacy Word mime — so the
 * mime alone misses them. Accept the real mime, a `.docx` extension on any
 * generic mime, or the zip signature (`PK\x03\x04`) on a Word / octet-stream
 * mime: an OLE `.doc` never starts with it, a `.docx` always does.
 */
export declare function isDocxMime(mime: string, filename?: string, data?: Buffer): boolean;
export declare function extractPdfText(data: Buffer): Promise<ExtractedText>;
export declare function extractDocxText(data: Buffer): ExtractedText;
/** The bytes of one entry of a zip archive, or undefined when the name is absent. */
export declare function readZipEntry(zip: Buffer, name: string): Buffer | undefined;
//# sourceMappingURL=document-text.d.ts.map