ODL PDF to Markdown

The #1 ranked open-source PDF parser. Convert any PDF to Markdown, JSON, and HTML with bounding-box coordinates for precise source citations in RAG pipelines.

Why ODL PDF?

Strong Points

• #1 in benchmarks — 0.90 overall extraction accuracy, 0.93 table accuracy across 200 real-world PDFs
• Bounding-box JSON — every element (heading, paragraph, table, image) has pixel coordinates for RAG citations
• Hybrid AI mode — routes complex pages to AI backend for charts, formulas, and borderless tables
• Built-in OCR — 80+ languages, works with scanned PDFs at 300 DPI+
• XY-Cut++ reading order — correct reading order for multi-column academic papers
• No API key — fully self-hosted, open-source Apache 2.0
• Multiple formats — Markdown (clean text), JSON (structured), HTML (rich layout)

• Markdown — clean readable text with correct reading order
• JSON — structured data with bounding boxes, font sizes, page numbers
• HTML — rich HTML output preserving layout
• OCR — built-in OCR for scanned PDFs (80+ languages)
• Tables — complex table extraction (0.93 accuracy)
• Reading order — XY-Cut++ algorithm for multi-column layouts
• No API key — fully self-hosted, open-source

Requirements

Java 11+ (symlink setup)

OpenDataLoader requires Java. After installing, create a symlink so Python subprocesses can find it:

# Find your Java install
ls ~/jdk-*/bin/java 2>/dev/null || ls /opt/jdk*/bin/java 2>/dev/null

# Create symlink
ln -sf /path/to/java/bin/java ~/.local/bin/java

Python 3.10+

pip install opendataloader-pdf

Or use the auto-install script (handles Java + Python automatically):

curl -fsSL https://raw.githubusercontent.com/opendataloader-project/opendataloader-pdf/main/scripts/install.sh | bash

Usage

CLI

# Basic — markdown + json output
pdf2md document.pdf ./output

# HTML + JSON output
pdf2md document.pdf ./output html,json

# Markdown only
pdf2md document.pdf ./output markdown

Python

import opendataloader_pdf

opendataloader_pdf.convert(
    input_path="document.pdf",
    output_dir="./output",
    format="markdown,json"
)

Supported Input Formats

Output Formats

Markdown

Clean text with heading hierarchy, bullet lists, and paragraph structure.

JSON

{
  "file name": "document.pdf",
  "number of pages": 5,
  "author": "Author Name",
  "kids": [
    {
      "type": "heading",
      "level": "Doctitle",
      "page number": 1,
      "bounding box": [100.0, 744.5, 404.0, 773.1],
      "font": "Helvetica-Bold",
      "font size": 24.0,
      "content": "Document Title"
    },
    {
      "type": "paragraph",
      "page number": 1,
      "bounding box": [100.0, 676.8, 316.3, 713.0],
      "font": "Helvetica",
      "font size": 14.0,
      "content": "Paragraph text..."
    }
  ]
}

Installation

OpenClaw Skill Install

clawhub install pdf-to-markdown

Manual Install

# Install dependencies
pip install opendataloader-pdf

# Make script executable
chmod +x scripts/pdf2md

Architecture

PDF Input
    │
    ▼
OpenDataLoader PDF (JVM)
    │
    ├── PDFBox    ──► Text extraction + layout analysis
    ├── veraPDF   ──► PDF validation + structure
    └── Tesseract ──► OCR (scanned PDFs)
    │
    ▼
Output: Markdown / JSON / HTML

Benchmark

OpenDataLoader ranked #1 against 9 other open-source and commercial PDF parsers on a test set of 200 real-world PDFs:

Common Use Cases

• RAG pipelines — convert PDFs to chunkable markdown
• Document parsing — extract text from research papers
• Accessibility — convert PDFs to structured data
• Data extraction — pull tables from reports
• Content migration — PDF to markdown for wikis/docs

See Also

• OpenDataLoader PDF — upstream project
• PDF/UA — PDF accessibility standard
• veraPDF — PDF/A validator