Extract (PDF Conversion)

PDFStract provides 10+ PDF extraction libraries to convert your documents into structured text ready for AI applications. Each library has different strengths, making it easy to choose the right tool for your specific documents.

Quick Start

from pdfstract import PDFStract

pdfstract = PDFStract()

# Auto-select best available library
text = pdfstract.convert('document.pdf', library='auto')

# Use specific library
text = pdfstract.convert('document.pdf', library='marker')

Available Libraries

By Installation Tier

Tier	Libraries	Install Command
Base	pymupdf4llm, markitdown	pip install pdfstract
Standard	+ pytesseract, unstructured	pip install pdfstract[standard]
Advanced	+ marker, docling, paddleocr, deepseek	pip install pdfstract[advanced]
All	All libraries combined	pip install pdfstract[all]

Library Comparison

Library	Speed	Quality	OCR	Best For
pymupdf4llm	⚡ Fast	Good	❌	Simple text PDFs, quick extraction
markitdown	⚡ Fast	Good	❌	Markdown output, Microsoft docs
marker	🔄 Medium	Excellent	✅	Complex layouts, figures, tables
docling	🔄 Medium	Excellent	✅	Document intelligence, structure
unstructured	🔄 Medium	Very Good	✅	Structured documents, forms
pytesseract	🐢 Slow	Good	✅	Scanned PDFs, legacy documents
paddleocr	🔄 Medium	Very Good	✅	Multi-language OCR
deepseek	⚡ Fast	Very Good	✅	GPU-accelerated OCR
mineru	🔄 Medium	Excellent	✅	Scientific papers, LaTeX

Python API

Basic Conversion

from pdfstract import PDFStract

pdfstract = PDFStract()

# Convert with auto-selection
text = pdfstract.convert('document.pdf', library='auto')

# Convert with specific library
text = pdfstract.convert('document.pdf', library='docling')

# Different output formats
markdown = pdfstract.convert('document.pdf', library='marker', output_format='markdown')
json_data = pdfstract.convert('document.pdf', library='docling', output_format='json')
plain_text = pdfstract.convert('document.pdf', library='pymupdf4llm', output_format='text')

Async Conversion

import asyncio
from pdfstract import PDFStract

async def process_pdf():
    pdfstract = PDFStract()
    result = await pdfstract.convert_async('document.pdf', library='marker')
    return result

text = asyncio.run(process_pdf())

Batch Conversion

from pdfstract import PDFStract

pdfstract = PDFStract()

# Process all PDFs in a directory
results = pdfstract.batch_convert(
    pdf_directory='./documents/',
    library='docling',
    output_format='markdown',
    parallel_workers=4
)

print(f"Success: {results['success']}, Failed: {results['failed']}")

for filename, content in results['results'].items():
    if 'error' not in content:
        print(f"✓ {filename}: {len(content)} chars")

List Available Libraries

from pdfstract import PDFStract

pdfstract = PDFStract()

# Get all libraries with status
libraries = pdfstract.list_libraries()
for lib in libraries:
    status = "✓" if lib['available'] else "✗"
    print(f"{status} {lib['name']}")

# Get only available library names
available = pdfstract.list_available_libraries()
print(f"Available: {available}")

CLI Usage

Convert Single PDF

# Auto-select best library
pdfstract convert document.pdf --library auto

# Use specific library
pdfstract convert document.pdf --library marker

# Save to file
pdfstract convert document.pdf --library docling --output result.md

# Different output formats
pdfstract convert document.pdf --library marker --format json
pdfstract convert document.pdf --library pymupdf4llm --format text

Batch Processing

# Process all PDFs in folder
pdfstract batch ./documents/ --output ./results/

# With parallel workers
pdfstract batch ./pdfs/ --library docling --parallel 4

# With specific output format
pdfstract batch ./docs/ --library marker --format json

Compare Libraries

# Compare multiple libraries on same document
pdfstract compare document.pdf --libraries marker,docling,pymupdf4llm

List Available Libraries

# Show all libraries with status
pdfstract libs

Output Formats

PDFStract supports multiple output formats:

Format	Description	Use Case
markdown	Markdown with headers, lists, tables	RAG, documentation
text	Plain text	Simple processing
json	Structured JSON with metadata	Programmatic access
html	HTML markup	Web display

# Markdown (default)
md = pdfstract.convert('doc.pdf', library='marker', output_format='markdown')

# JSON with structure
data = pdfstract.convert('doc.pdf', library='docling', output_format='json')

# Plain text
text = pdfstract.convert('doc.pdf', library='pymupdf4llm', output_format='text')

Library Details

pymupdf4llm

Fast, lightweight extraction using PyMuPDF. Best for simple text PDFs.

text = pdfstract.convert('simple.pdf', library='pymupdf4llm')

Pros: Very fast, low memory, no GPU needed
Cons: No OCR, struggles with complex layouts

marker

ML-powered extraction with excellent quality. Handles complex layouts, figures, and tables.

text = pdfstract.convert('complex.pdf', library='marker')

Pros: Excellent quality, handles tables/figures, OCR support
Cons: Requires model download, slower than basic extractors

docling

IBM's document intelligence platform. Great for structured documents.

text = pdfstract.convert('report.pdf', library='docling')

Pros: Document structure understanding, metadata extraction
Cons: Requires model download, heavier dependencies

unstructured

Smart document parsing with element detection.

text = pdfstract.convert('form.pdf', library='unstructured')

Pros: Element detection, form handling, structure preservation
Cons: Moderate speed, requires additional dependencies

paddleocr

Accurate OCR engine with multi-language support.

text = pdfstract.convert('scanned.pdf', library='paddleocr')

Pros: Multi-language, accurate OCR, GPU acceleration
Cons: Requires PaddlePaddle installation

Model Downloads

Some libraries require model downloads before first use:

# Download models for specific library
pdfstract download marker
pdfstract download docling

# Download all available models
pdfstract download --all

# Check download status
pdfstract = PDFStract()
status = pdfstract.get_converter_status('marker')
print(f"Download status: {status['download_status']}")

# Prepare/download models
result = pdfstract.prepare_converter('marker')
print(f"Ready: {result['success']}")

Error Handling

from pdfstract import PDFStract

pdfstract = PDFStract()

try:
    text = pdfstract.convert('document.pdf', library='marker')
except FileNotFoundError:
    print("PDF file not found")
except ValueError as e:
    print(f"Conversion error: {e}")

Best Practices

1. Use auto for portability: Your code works regardless of installed tier
2. Match library to document type: Use OCR libraries for scanned PDFs
3. Batch process for efficiency: Amortizes library loading time
4. Download models in advance: Avoid first-run delays in production

Next Steps

• Chunk - Split extracted text for RAG
• Embed - Generate vector embeddings
• CLI Guide - Full command reference
• Python API - Complete API documentation