opendataloader-pdf-llamaindex

LlamaIndex reader for OpenDataLoader PDF — parse PDFs into structured Document objects for RAG pipelines.

For the full feature set of the core engine (hybrid AI mode, OCR, formula extraction, benchmarks, accessibility), see the OpenDataLoader PDF documentation.

Features

• Accurate reading order — XY-Cut++ algorithm handles multi-column layouts correctly
• Table extraction — Preserves table structure in output
• Multiple formats — Text, Markdown, JSON (with bounding boxes), HTML
• Per-page splitting — Each page becomes a separate Document with page number metadata
• AI safety — Built-in prompt injection filtering (hidden text, off-page content, invisible layers)
• 100% local — No cloud APIs, your documents never leave your machine
• Fast — Rule-based extraction, no GPU required

Requirements

• Python >= 3.10
• Java 11+ available on system PATH

Verify Java is installed:

java -version

Installation

pip install -U opendataloader-pdf-llamaindex

Quick Start

from llama_index.readers.opendataloader_pdf import OpenDataLoaderPDFReader

reader = OpenDataLoaderPDFReader(format="text")
documents = reader.load_data(file_path="document.pdf")

print(documents[0].text)
print(documents[0].metadata)
# {'source': 'document.pdf', 'format': 'text', 'page': 1}

SimpleDirectoryReader Integration

Use with LlamaIndex's SimpleDirectoryReader via the file_extractor parameter:

from llama_index.core import SimpleDirectoryReader
from llama_index.readers.opendataloader_pdf import OpenDataLoaderPDFReader

reader = SimpleDirectoryReader(
    input_dir="./documents",
    file_extractor={".pdf": OpenDataLoaderPDFReader(format="markdown")}
)
documents = reader.load_data()

Usage Examples

Output Formats

from llama_index.readers.opendataloader_pdf import OpenDataLoaderPDFReader

# Plain text (default) — best for simple RAG
reader = OpenDataLoaderPDFReader(format="text")

# Markdown — preserves headings, lists, tables
reader = OpenDataLoaderPDFReader(format="markdown")

# JSON — structured data with bounding boxes for source citations
reader = OpenDataLoaderPDFReader(format="json")

# HTML — styled output
reader = OpenDataLoaderPDFReader(format="html")

Tagged PDF Support

For accessible PDFs with structure tags (common in government/legal documents):

reader = OpenDataLoaderPDFReader(use_struct_tree=True)

Table Detection

reader = OpenDataLoaderPDFReader(
    format="markdown",
    table_method="cluster"  # Better for borderless tables
)

Sensitive Data Sanitization

reader = OpenDataLoaderPDFReader(sanitize=True)
# Replaces emails, phone numbers, IPs, credit cards, URLs with placeholders

Page Selection

reader = OpenDataLoaderPDFReader(pages="1,3,5-7")

Headers and Footers

reader = OpenDataLoaderPDFReader(include_header_footer=True)

Password-Protected PDFs

reader = OpenDataLoaderPDFReader(password="secret")

Image Handling

# Embed images as Base64 in output
reader = OpenDataLoaderPDFReader(image_output="embedded")

# Save images to external files
reader = OpenDataLoaderPDFReader(
    image_output="external",
    image_dir="./extracted_images"
)

Hybrid AI Mode

For higher accuracy on complex documents (requires a running hybrid backend):

reader = OpenDataLoaderPDFReader(
    hybrid="docling-fast",
    hybrid_fallback=True  # Fall back to Java on backend failure
)

RAG Pipeline Example

from llama_index.core import VectorStoreIndex, SimpleDirectoryReader
from llama_index.readers.opendataloader_pdf import OpenDataLoaderPDFReader

# Load PDFs
reader = SimpleDirectoryReader(
    input_dir="./documents",
    file_extractor={".pdf": OpenDataLoaderPDFReader(format="markdown")}
)
documents = reader.load_data()

# Build index and query
index = VectorStoreIndex.from_documents(documents)
query_engine = index.as_query_engine()
response = query_engine.query("What are the key findings?")
print(response)

Parameters

Document Metadata

Each Document includes metadata:

With split_pages=True (default):

{"source": "document.pdf", "format": "text", "page": 1}

With split_pages=False:

{"source": "document.pdf", "format": "text"}

With hybrid mode:

{"source": "document.pdf", "format": "text", "page": 1, "hybrid": "docling-fast"}