pdfstract 1.0.1

PDFStract - Unified PDF Extraction & Conversion CLI + Web UI with 10+ extraction libraries

PDFStract CLI - Command-Line Interface Guide

PDFStract now includes a powerful command-line interface for PDF extraction and conversion with support for batch processing, multi-library comparison, and production automation.

Table of Contents

• Installation
• Quick Start
• Commands
• Batch Processing
• Real-World Examples
• Integration Examples
• Performance Tips
• Troubleshooting

Installation

From PyPI (Recommended)

pip install pdfstract

From Source

git clone https://github.com/aksarav/pdfstract.git
cd pdfstract
pip install -e .
# or
uv sync

Verify Installation

pdfstract --help

Quick Start

1. List Available Libraries

pdfstract libs

Output shows which PDF extraction libraries are installed:

✓ Available: unstructured, marker, pymupdf4llm, docling, ...
✗ Unavailable: (with error reasons)

2. Convert a Single PDF

# Convert to markdown (default)
pdfstract convert document.pdf --library unstructured

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

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

3. Test Multiple Libraries

pdfstract compare sample.pdf \
  -l unstructured \
  -l marker \
  -l pymupdf4llm \
  --format markdown \
  --output ./comparison_results

Results in:

comparison_results/
├─ unstructured_result.md
├─ marker_result.md
├─ pymupdf4llm_result.md
└─ comparison_summary.json

Commands

pdfstract libs

List all available PDF extraction libraries and their status.

pdfstract libs

Output:

• Shows 10+ libraries (PyMuPDF4LLM, MarkItDown, Marker, Docling, etc.)
• Displays availability status (✓ Available / ✗ Unavailable)
• Shows error messages for unavailable libraries

---

pdfstract convert

Convert a single PDF file with a specified library.

pdfstract convert INPUT_FILE [OPTIONS]

Options:

• -l, --library TEXT (required) - Extraction library to use
• -f, --format [markdown|json|text] - Output format (default: markdown)
• -o, --output PATH - Output file path (optional, prints to stdout if not specified)

Examples:

# Print to terminal
pdfstract convert sample.pdf --library unstructured

# Save to file
pdfstract convert sample.pdf --library marker --output result.md

# JSON format
pdfstract convert sample.pdf --library docling --format json --output result.json

---

pdfstract compare

Compare multiple extraction libraries on a single PDF to find the best one.

pdfstract compare INPUT_FILE [OPTIONS]

Options:

• -l, --libraries TEXT (required, multiple) - Libraries to compare
• -f, --format [markdown|json|text] - Output format (default: markdown)
• -o, --output PATH (required) - Output directory for results

Examples:

# Compare 3 libraries
pdfstract compare sample.pdf \
  -l unstructured \
  -l marker \
  -l pymupdf4llm \
  --output ./test_results

# Compare with JSON output
pdfstract compare invoice.pdf \
  -l marker \
  -l docling \
  --format json \
  --output ./compare

Output:

• Individual result files for each library
• comparison_summary.json with metadata and stats

---

pdfstract batch

Batch convert multiple PDFs in a directory with parallel processing.

pdfstract batch INPUT_DIRECTORY [OPTIONS]

Options:

• -l, --library TEXT (required) - Extraction library to use
• -f, --format [markdown|json|text] - Output format (default: markdown)
• -o, --output PATH (required) - Output directory
• -p, --parallel INTEGER - Number of parallel workers (default: 2)
• --pattern TEXT - File pattern to match (default: *.pdf)
• --skip-errors - Skip PDFs that fail conversion

Examples:

# Basic batch conversion
pdfstract batch ./documents \
  --library unstructured \
  --output ./converted

# With parallel processing
pdfstract batch ./documents \
  --library marker \
  --output ./converted \
  --parallel 4

# With error handling
pdfstract batch ./pdfs \
  --library docling \
  --format json \
  --output ./converted \
  --parallel 8 \
  --skip-errors

# Custom file pattern
pdfstract batch ./invoices \
  --library unstructured \
  --pattern "invoice_*.pdf" \
  --output ./structured

Output:

output_directory/
├─ file1.md
├─ file2.md
├─ file3.md
├─ ... (more files)
└─ batch_report.json

Batch Report (batch_report.json):

{
  "input_directory": "/path/to/pdfs",
  "output_directory": "/path/to/output",
  "library": "unstructured",
  "format": "markdown",
  "total_files": 150,
  "statistics": {
    "success": 147,
    "failed": 2,
    "skipped": 1
  },
  "files": {
    "document1.pdf": {
      "status": "success",
      "size_bytes": 45230
    },
    "document2.pdf": {
      "status": "failed",
      "error": "Invalid PDF format"
    }
  }
}

---

pdfstract batch-compare

Compare multiple extraction libraries across an entire corpus of PDFs.

pdfstract batch-compare INPUT_DIRECTORY [OPTIONS]

Options:

• -l, --libraries TEXT (required, multiple) - Libraries to compare
• -f, --format [markdown|json|text] - Output format (default: markdown)
• -o, --output PATH (required) - Output directory
• --max-files INTEGER - Limit number of files to process

Examples:

# Compare on all PDFs
pdfstract batch-compare ./papers \
  -l marker \
  -l unstructured \
  -l pymupdf4llm \
  --output ./library_comparison

# Quick test on sample
pdfstract batch-compare ./large_corpus \
  -l marker \
  -l unstructured \
  --max-files 50 \
  --output ./sample_test

Output:

• batch_comparison_report.json with per-library success rates
• Per-file results for all PDFs tested

Batch Processing

When to Use Batch Processing

Batch processing is perfect for:

• Converting 100+ PDFs with one library
• Testing multiple libraries on entire corpus
• Production automation jobs
• Legacy archive digitization
• Enterprise migrations

Parallel Processing Guidelines

Choose workers based on library and hardware:

Library	CPU Usage	Recommended Workers
PyMuPDF4LLM	Low	8-16
MarkItDown	Medium	4-8
Unstructured	Medium	4-6
Marker (ML)	High	2-4
OCR (Paddle/Tesseract)	Very High	1-2

# Fast library, beefy server
pdfstract batch ./docs --library pymupdf4llm --parallel 16

# Slow ML library
pdfstract batch ./docs --library marker --parallel 2

# Medium library, balanced
pdfstract batch ./docs --library unstructured --parallel 6

Error Handling

Without --skip-errors (default):

• Stops on first error
• Exit code 1 if failures occur
• Best for strict pipelines

With --skip-errors:

• Continues processing all files
• Failed files marked in report
• Exit code 0 (always succeeds)
• Best for best-effort processing

# Strict mode (fail on errors)
pdfstract batch ./docs --library unstructured --output ./result

# Best-effort mode (skip errors)
pdfstract batch ./docs --library unstructured --output ./result --skip-errors

Parsing Batch Reports

Use jq to analyze batch reports:

# Overall statistics
jq '.statistics' batch_report.json

# Success rate
jq '.statistics.success / .total_files * 100' batch_report.json

# Failed files only
jq '.files | to_entries[] | select(.value.status=="failed")' batch_report.json

# Average output size
jq '.files | to_entries[] | select(.value.status=="success") | .value.size_bytes' batch_report.json | \
  awk '{sum+=$1; count++} END {print sum/count/1024 " KB"}'

Real-World Examples

Example 1: Law Firm Document Digitization

Scenario: Convert 5,000 case files to searchable markdown

# Step 1: Test on sample (5 cases)
pdfstract compare case_1.pdf case_2.pdf case_3.pdf \
  -l marker \
  -l unstructured \
  -l docling \
  --output ./test_results

# Step 2: Review outputs, pick best library (e.g., docling)

# Step 3: Full batch conversion (5,000 cases)
pdfstract batch ./all_cases \
  --library docling \
  --format markdown \
  --output ./converted_cases \
  --parallel 8 \
  --skip-errors

# Step 4: Monitor results
jq '.statistics' ./converted_cases/batch_report.json

Results: 2 months manual work → 12 hours automated. $100k labor cost → $500 compute cost.

---

Example 2: Research Paper Quality Testing

Scenario: Find best extractor for 1,000 research papers

# Test on sample
pdfstract batch-compare ./papers \
  -l marker \
  -l unstructured \
  -l pymupdf4llm \
  --max-files 50 \
  --output ./library_test

# Review success rates, pick best

# Full batch with chosen library
pdfstract batch ./papers \
  --library marker \
  --format json \
  --output ./extracted_papers \
  --parallel 4

---

Example 3: Invoice Processing Pipeline

Scenario: Daily automated invoice conversion

# Create batch scheduler config
pdfstract-scheduler create daily_invoices \
  ./daily_invoices_input \
  ./daily_invoices_output \
  --library unstructured \
  --parallel 4

# Run job
pdfstract-scheduler run daily_invoices

# View results
cat ./daily_invoices_output/batch_report.json

---

Example 4: Legacy Archive Migration

Scenario: Modernize 50,000 legacy PDFs to JSON

pdfstract batch ./legacy_archive \
  --library marker \
  --format json \
  --output ./modern_archive \
  --parallel 16 \
  --skip-errors

# Monitor with tail
tail -f ./modern_archive/batch_report.json

Integration Examples

Bash Script: Nightly Batch Job

#!/bin/bash
DATE=$(date +%Y%m%d)
OUTPUT_DIR="./converted/$DATE"

pdfstract batch ./daily_pdfs \
  --library unstructured \
  --format markdown \
  --output "$OUTPUT_DIR" \
  --parallel 8 \
  --skip-errors

# Alert if failures
FAILED=$(jq '.statistics.failed' "$OUTPUT_DIR/batch_report.json")
if [ "$FAILED" -gt 0 ]; then
  echo "⚠️  $FAILED conversions failed on $DATE" | mail [e-mail]
fi

---

Python: Programmatic Usage

import subprocess
import json
from pathlib import Path

def batch_convert(pdf_dir: str, library: str, output_dir: str) -> dict:
    """Wrapper around CLI"""
    result = subprocess.run([
        'pdfstract', 'batch', pdf_dir,
        '--library', library,
        '--output', output_dir,
        '--parallel', '4'
    ])
    
    # Load and parse report
    report_file = Path(output_dir) / 'batch_report.json'
    with open(report_file) as f:
        report = json.load(f)
    
    success_rate = (report['statistics']['success'] / 
                   report['total_files'] * 100)
    print(f"Success Rate: {success_rate:.1f}%")
    return report

---

Docker: Containerized Processing

FROM python:3.13-slim
RUN pip install pdfstract

ENTRYPOINT ["pdfstract"]
CMD ["batch", "/data/input", "--library", "unstructured", "--output", "/data/output"]

# Build
docker build -t pdfstract .

# Run
docker run -v ./pdfs:/data/input -v ./converted:/data/output pdfstract

---

CI/CD: GitHub Actions Example

name: PDF Extraction

on: [push]

jobs:
  extract:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v2
      
      - name: Install PDFStract
        run: pip install pdfstract
      
      - name: Extract PDFs
        run: |
          pdfstract batch ./source_pdfs \
            --library unstructured \
            --format json \
            --output ./extracted
      
      - name: Upload results
        uses: actions/upload-artifact@v2
        with:
          name: extracted-pdfs
          path: extracted/

Performance Tips

1. Test Before Large Batches

# Always test library first
pdfstract convert sample.pdf --library CHOSEN_LIB

# Then run batch
pdfstract batch ./1000_files --library CHOSEN_LIB --parallel 4

2. Choose Library Based on Speed vs Quality

• Speed: pymupdf4llm (--parallel 16)
• Balanced: unstructured (--parallel 6)
• Quality: marker (--parallel 2)

3. Monitor Long-Running Jobs

pdfstract batch ./files --library marker --parallel 2 2>&1 | tee job.log

# In another terminal
tail -f job.log

4. Retry Failed Conversions

# Extract failed files from report
jq -r '.files | to_entries[] | select(.value.status=="failed") | .key' \
  batch_report.json > failed.txt

# Retry with different library
while read file; do
  pdfstract convert "$file" --library marker --output "retry_results/${file%.pdf}.md"
done < failed.txt

Troubleshooting

"Library 'X' not available"

Solution:

pdfstract libs  # See what's available

# Install missing library
uv add LIBRARY_NAME
# or
pip install LIBRARY_NAME

"File is not a PDF"

Only PDF files are supported. Check:

• File extension is .pdf
• File is actually a PDF (not renamed)
• File is not corrupted

Batch Job Very Slow

Reduce parallel workers:

# Instead of --parallel 8
pdfstract batch ./docs --library marker --parallel 2

Or distribute job across multiple machines.

Memory Running Out

Reduce workers or process in smaller batches:

pdfstract batch ./docs --library marker --parallel 1

DeepSeek-OCR Not Working

Requires CUDA GPU. Alternatives:

• PaddleOCR (no GPU needed)
• Pytesseract (no GPU needed)
• Unstructured (CPU-based)

Batch Job Scheduler

For recurring jobs, use the batch scheduler:

# Create scheduled job
pdfstract-scheduler create daily_job \
  ./input_dir \
  ./output_dir \
  --library unstructured \
  --parallel 4

# Run job
pdfstract-scheduler run daily_job

# View execution history
pdfstract-scheduler history daily_job

# List all jobs
pdfstract-scheduler list

Add to cron for automated scheduling:

0 2 * * * pdfstract-scheduler run daily_job

Next Steps

• Read main README.md for Web UI guide
• Check batch processing use cases
• Join GitHub discussions for questions
• Open issues for bugs or feature requests

Support

• GitHub: https://github.com/aksarav/pdfstract
• Issues: https://github.com/aksarav/pdfstract/issues
• Discussions: https://github.com/aksarav/pdfstract/discussions

---

Happy extracting! 🚀📄✨