llama-index-readers-pdfmux 0.1.0

LlamaIndex reader for pdfmux -- self-healing PDF extraction for RAG pipelines

llama-index-readers-pdfmux

LlamaIndex reader for pdfmux -- self-healing PDF extraction for RAG pipelines.

Why pdfmux?

Most PDF loaders use a single extraction method and silently fail on complex layouts. pdfmux routes each page through the best extraction pipeline automatically:

• Smart routing -- selects the optimal parser per page (text-heavy, scanned, tables, mixed)
• Confidence scoring -- every chunk includes a confidence score so your RAG pipeline can filter or re-rank
• Self-healing -- retries with alternative extractors when the primary one returns low-quality output

Install

pip install llama-index-readers-pdfmux

Usage

from llama_index_readers_pdfmux import PDFMuxReader

reader = PDFMuxReader()
docs = reader.load_data("report.pdf")

Each Document includes metadata with extraction quality signals:

reader = PDFMuxReader(quality="high")
for doc in reader.load_data("report.pdf"):
    print(doc.metadata)
    # {
    #   "source": "report.pdf",
    #   "title": "Q4 Results",
    #   "page_start": 1,
    #   "page_end": 3,
    #   "tokens": 820,
    #   "confidence": 0.94
    # }

Options

# Quality presets: "fast", "standard" (default), "high"
reader = PDFMuxReader(quality="high")

# Load all PDFs in a directory
docs = reader.load_data("./papers/")

# Custom glob pattern
reader = PDFMuxReader(glob="**/*.pdf")
docs = reader.load_data("./papers/")

# Attach extra metadata
docs = reader.load_data("report.pdf", extra_info={"project": "Q4 analysis"})

With LlamaIndex pipelines

from llama_index.core import VectorStoreIndex
from llama_index_readers_pdfmux import PDFMuxReader

reader = PDFMuxReader(quality="high")
docs = reader.load_data("./papers/")

# Filter low-confidence chunks
docs = [d for d in docs if d.metadata["confidence"] > 0.8]

index = VectorStoreIndex.from_documents(docs)
query_engine = index.as_query_engine()
response = query_engine.query("What were the key findings?")

License

MIT