import { LoadPdfOptions, RagLoaderResult } from "./loader.type.mjs";

//#region ../ai/src/rag/loaders/load-pdf.d.ts
/**
 * Load a PDF's bytes into {@link RagDocument}(s) via the OPTIONAL `pdf-parse`
 * peer. The peer is resolved lazily on the FIRST call (not at import) so
 * importing `@warlock.js/ai` never forces it to be installed; when it is
 * absent the curated {@link PDF_PARSE_INSTALL_INSTRUCTIONS} is thrown as a
 * plain `Error` (a missing optional peer is an infrastructure fault, not a
 * content problem).
 *
 * By default the whole PDF becomes a single document carrying
 * `metadata.pageCount`. With `perPage: true`, each page becomes its own
 * document (`id` suffixed `#p<n>`, `metadata.page` set) so citations stay
 * page-precise. Document `metadata.title` comes from the PDF info
 * dictionary's `Title` (unless overridden), and `metadata.loader` is
 * `"pdf"`. The output is the exact shape `index()` consumes.
 *
 * @example
 * import { readFile } from "node:fs/promises";
 * const kb = ai.rag({ embedder, store });
 * await kb.index(await loadPdf(await readFile("guide.pdf"), { id: "guide" }));
 *
 * @example
 * // One document per page for page-precise citations:
 * await kb.index(await loadPdf(bytes, { id: "manual", perPage: true }));
 *
 * @param input - The PDF bytes (`Buffer`, `ArrayBuffer`, or `Uint8Array`).
 * @param options - `perPage` plus shared `id` / `metadata` / `tags`
 *   ({@link LoadPdfOptions}).
 * @returns A {@link RagLoaderResult} ready for `rag.index()`.
 * @throws {Error} carrying {@link PDF_PARSE_INSTALL_INSTRUCTIONS} when the
 *   `pdf-parse` peer is not installed.
 */
declare function loadPdf(input: Buffer | ArrayBuffer | Uint8Array, options?: LoadPdfOptions): Promise<RagLoaderResult>;
//#endregion
export { loadPdf };
//# sourceMappingURL=load-pdf.d.mts.map