pdffr-llamaindex

A LlamaIndex.TS reader backed by pdffr: each PDF page becomes a Document of Markdown (headings, lists, tables, math), decompiled from the file's own glyph geometry in milliseconds, with on-device OCR only where a page needs it. Local and private — nothing is uploaded.

npm install pdffr-llamaindex @llamaindex/core

import { PdffrReader } from 'pdffr-llamaindex';

const reader = new PdffrReader({ lang: 'eng' });
const docs = await reader.loadData('report.pdf');
// docs[i].text     → markdown for page i+1
// docs[i].metadata → { file_path, page, totalPages, ocrRegions, nativeChars }

Works with SimpleDirectoryReader via fileExtToReader: { pdf: new PdffrReader() }. Options: ocr (default true), lang, splitPages (default true), concurrency.

MIT © Amer Sarhan