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

Parameter	Type	Default	Description
format	str	"text"	Output format: "text", "markdown", "json", "html"
split_pages	bool	True	Split output into separate Documents per page
quiet	bool	False	Suppress CLI logging output
content_safety_off	list[str]	None	Safety filters to disable: "all", "hidden-text", "off-page", "tiny", "hidden-ocg"
password	str	None	Password for encrypted PDFs
keep_line_breaks	bool	False	Preserve original line breaks
replace_invalid_chars	str	None	Replacement for unrecognized characters
use_struct_tree	bool	False	Use PDF structure tree (tagged PDFs)
table_method	str	None	"default" (border-based) or "cluster" (border + cluster)
reading_order	str	None	"off" or "xycut" (default when not specified)
image_output	str	"off"	"off", "embedded" (Base64), "external" (files)
image_format	str	None	"png" or "jpeg"
image_dir	str	None	Directory for external images
sanitize	bool	False	Mask emails, phones, IPs, credit cards, URLs
pages	str	None	Pages to extract, e.g., "1,3,5-7"
include_header_footer	bool	False	Include page headers and footers
detect_strikethrough	bool	False	Detect strikethrough text (experimental)
hybrid	str	None	Hybrid AI backend: "docling-fast"
hybrid_mode	str	None	"auto" (complex pages only) or "full" (all pages)
hybrid_url	str	None	Custom backend server URL
hybrid_timeout	str	None	Backend timeout in milliseconds
hybrid_fallback	bool	False	Fall back to Java on backend failure

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"}