pdf2llm - PDF to LLM Context Extractor

Extract content from PDFs in a format optimized for Large Language Model (LLM) consumption.

Features

• Multiple formats: Extract as Markdown (preserves structure) or plain text
• Image extraction: Automatically extracts and saves images with configurable DPI
• Table preservation: Maintains table structure in Markdown format
• Page boundaries: Optional page markers for maintaining document structure
• Batch processing: Process multiple PDFs at once
• Organized output: Clean directory structure for each PDF
• Structure analysis: Analyze PDFs before extraction
• Token estimation: Get token counts for LLM context planning

Installation

# Install from PyPI
pip install pdf2llm
# or
uv pip install pdf2llm

# Or install from source
git clone https://github.com/yourusername/pdf2llm.git
cd pdf2llm
uv sync

Usage

Command Line Interface

# Basic extraction
uv run ./pdf2llm document.pdf

# Extract to specific directory
uv run ./pdf2llm document.pdf -o extracted_docs/

# Batch process multiple PDFs
uv run ./pdf2llm *.pdf -o zoning_docs/

# Extract as plain text without images
uv run ./pdf2llm document.pdf --format text --no-images

# Analyze PDF structure only
uv run ./pdf2llm document.pdf --analyze-only

# High quality image extraction
uv run ./pdf2llm document.pdf --dpi 300

# Get JSON output for integration
uv run ./pdf2llm document.pdf --json

# Set token limit warning
uv run ./pdf2llm document.pdf --token-limit 4000

Python API

from pdf_utils import PDFExtractor

# Create extractor
extractor = PDFExtractor(
    output_dir=Path("extracted"),
    image_format="png",
    dpi=150
)

# Extract single PDF
result = extractor.extract(
    Path("document.pdf"),
    output_format="markdown"
)

print(f"Tokens: {result.token_estimate}")
print(f"Pages: {result.page_count}")
print(f"Has images: {result.has_images}")
print(f"Has tables: {result.has_tables}")

# Save to file
output_path = extractor.save_extraction(result, Path("document.pdf"))

# Batch extraction
pdf_files = list(Path("pdfs/").glob("*.pdf"))
results = extractor.batch_extract(pdf_files)

Output Structure

extracted/
├── document_name/
│   ├── content.md         # Extracted content
│   └── images/           # Extracted images (if any)
│       ├── page-1-0.png
│       └── page-2-0.png
└── another_document/
    ├── content.md
    └── images/

Use Cases

Zoning Documents Analysis

# Extract all zoning PDFs with high-quality images
uv run ./pdf2llm zoning_*.pdf -o zoning_analysis/ --dpi 300

# Then in your Python code:
with open("zoning_analysis/zoning_code_2024/content.md", "r") as f:
    content = f.read()

# Use with your LLM
response = llm.chat(
    messages=[{
        "role": "system",
        "content": f"You are analyzing zoning documents. Document: {content}"
    }, {
        "role": "user", 
        "content": "What are the setback requirements for R-1 zones?"
    }]
)

Document Q&A System

# Process all documents
uv run ./pdf2llm documents/*.pdf -o knowledge_base/

# Check token counts
uv run ./pdf2llm documents/*.pdf --json | jq '.token_estimate'

Research Paper Analysis

# Extract with tables and figures
uv run ./pdf2llm research_paper.pdf --dpi 200

# Extract text only for quick analysis
uv run ./pdf2llm research_paper.pdf --format text --no-images

CLI Options

Package Structure

pdf_utils/
├── core/
│   └── extractor.py      # Core extraction logic
├── cli/
│   └── main.py          # CLI interface
└── __init__.py         # Package exports

Requirements

• Python 3.12+
• uv (for dependency management)
• Dependencies managed in pyproject.toml

License

MIT License