pdfsmith

PDF to Markdown conversion with multiple backend support

A unified interface to 19+ PDF parsing libraries including frontier LLMs. Pick the right tool for the job, or let pdfsmith choose for you.

Why pdfsmith?

• One API, many backends - Switch between parsers without changing your code
• Auto-selection - Automatically uses the best available parser
• Lightweight core - Install only the backends you need
• Battle-tested - Wrappers refined through extensive benchmarking

Installation

# Core package (no backends)
pip install pdfsmith

# With lightweight backends
pip install pdfsmith[light]

# Recommended stack (good balance of quality and speed)
pip install pdfsmith[recommended]

# All open-source backends
pip install pdfsmith[all]

# Frontier LLMs (GPT, Claude, Gemini)
pip install pdfsmith[frontier]

# Commercial cloud APIs
pip install pdfsmith[commercial]

# Specific backend
pip install pdfsmith[docling]

Quick Start

from pdfsmith import parse

# Auto-select best available backend
markdown = parse("document.pdf")

# Use a specific backend
markdown = parse("document.pdf", backend="docling")

# Check available backends
from pdfsmith import available_backends
for backend in available_backends():
    print(f"{backend.name}: {backend.description}")

CLI Usage

# Parse PDF to stdout
pdfsmith parse document.pdf

# Parse to file
pdfsmith parse document.pdf -o output.md

# Use specific backend
pdfsmith parse document.pdf -b docling

# List available backends
pdfsmith backends

Available Backends

Open Source

Commercial Cloud APIs

Frontier LLMs

Note: Frontier LLM backends require API keys set via environment variables (ANTHROPIC_API_KEY, OPENAI_API_KEY, GOOGLE_API_KEY).

Choosing a Backend

• Best quality: anthropic or openai - Frontier LLM accuracy (highest cost)
• Best value: llamaparse - Near-LLM quality at 10x lower cost
• Structure preservation: docling - Deep learning, GPU recommended
• Academic papers: marker - Optimized for LaTeX/equations
• Tables: pdfplumber - Excellent table detection
• Speed: pymupdf or kreuzberg - Fast extraction
• Minimal dependencies: pypdf - Pure Python, no binaries
• Budget LLM: gemini with gemini-2.0-flash - Very low cost LLM option

System Dependencies

Some backends require system packages for OCR functionality:

Tesseract OCR (for kreuzberg and unstructured with OCR):

# Ubuntu/Debian
sudo apt-get install tesseract-ocr

# macOS
brew install tesseract

# Windows
# Download from https://github.com/UB-Mannheim/tesseract/wiki

Without tesseract, these backends will still work for text-based PDFs but cannot extract text from scanned/image PDFs.

Async Support

from pdfsmith import parse_async

# Async parsing (uses backend's native async if available)
markdown = await parse_async("document.pdf")

Benchmarks

pdfsmith's backend wrappers were developed and refined through the pdf-bench benchmarking project, which evaluates parser performance across diverse document types.

License

MIT

Contributing

Contributions welcome! Please read our contributing guidelines before submitting PRs.