A CLI tool for extracting text from various document formats with configurable OCR capabilities.

• Multi-format Support: PDF, Word, HTML, E-books, Images, and text files
• Configurable OCR: llm-caller (Call LLM models) or surya_ocr (local OCR engine)
• Smart Content Strategy: Choose between text-first or image-first processing for PDFs
• Interactive Tool Selection: Auto-detects available tools and prompts for selection
• Automatic Tool Detection: No configuration files needed - tools are detected when required
• Cross-platform: macOS, Linux, Windows with automatic tool detection

• 🚀 Quick Start - Installation and basic usage
• 🔧 Development - Architecture and development guide
• 🏷️ Versioning - Version management and build system

# macOS
brew install ghostscript pandoc calibre && pip install surya-ocr

# Ubuntu/Linux  
sudo apt-get install ghostscript pandoc calibre && pip install surya-ocr

# Windows (with Chocolatey)
choco install ghostscript pandoc calibre && pip install surya-ocr

Download pre-built binaries from Releases or build from source with go build.

# Extract with interactive selection (prompts for OCR tool and content type)
doc-to-text document.pdf

# Use specific OCR tool
doc-to-text document.pdf --ocr surya_ocr
doc-to-text document.pdf --ocr llm-caller --llm-template qwen-vl-ocr

# Specify content processing strategy for PDFs
doc-to-text document.pdf --content-type text    # Try Calibre first, OCR fallback
doc-to-text document.pdf --content-type image   # Direct OCR processing

# Custom output
doc-to-text document.pdf -o output.txt

# Display and set language
doc-to-text language
# Switch language
DOC_TEXT_LANG=zh doc-to-text -V

# Version info
doc-to-text --version    # Quick version
doc-to-text version      # Detailed build info

The tool automatically detects required tools when needed:

• OCR Tools: llm-caller, surya_ocr
• Document Processing: ebook-convert (Calibre), pandoc, gs (Ghostscript)
• Detection Strategy: Command lookup → Common paths → Clear error messages

# Temporarily override settings
DOC_TEXT_OCR_STRATEGY=surya_ocr doc-to-text document.pdf
DOC_TEXT_CONTENT_TYPE=text doc-to-text document.pdf
DOC_TEXT_MAX_CONCURRENCY=8 doc-to-text document.pdf

• Local and multilingual (100+ languages)
• Installation: pip install surya-ocr
• Best for: Standard documents, batch processing

• AI-powered with template-based approach
• Requires: --llm-template parameter
• Best for: Complex layouts, handwritten text, specific models

• Default mode: Automatically prompts for tool selection
• Smart detection: Shows only available engines
• Auto-selection: For text content-type, automatically selects best tool without prompts

The --content-type parameter determines PDF processing strategy:

• text: Tries Calibre first (fast for text-based PDFs), then OCR if failed
• image: Uses OCR directly (default, best for scanned documents)

Text is extracted to organized directories:

• Input: /path/to/document.pdf
• Output: /path/to/{md5_hash}/text.txt
• Pages: /path/to/{md5_hash}/pages/ (for PDFs)

Large document processing can be interrupted and resumed. The tool automatically:

• Detects completed pages and skips them
• Continues from the last processed page
• Maintains processing state in intermediate directories

OCR tool not found: Tools are automatically detected. Ensure they are installed and available in your PATH

Permission errors: Ensure tools are executable and paths are accessible

Poor OCR quality: Try different OCR engines or ensure good source quality (300 DPI recommended)

MIT License