LlamaIndex Readers Integration: pdfmux

pdfmux extracts PDFs into section-aware, LLM-ready chunks. It scores each page's extraction confidence and routes low-confidence or scanned pages to an OCR fallback automatically, so mixed digital/scanned documents come back as clean text without manual pre-processing. Each chunk is returned as a LlamaIndex Document with title, page range, token estimate, and confidence metadata.

Installation

pip install llama-index-readers-pdfmux

Usage

from llama_index.readers.pdfmux import PDFMuxReader

reader = PDFMuxReader(quality="standard")  # "fast" | "standard" | "high"
documents = reader.load_data("report.pdf")

for doc in documents:
    print(doc.metadata["title"], "—", doc.metadata["tokens"], "tokens")

Each returned Document carries metadata: source, title, page_start, page_end, tokens, confidence. Pass extra_info={...} to load_data to merge additional metadata into every document.

This loader is designed for RAG/LLM pipelines that ingest a mix of scanned and digital PDFs and need per-page confidence to decide what to trust.