kreuzberg-dev/kreuzberg v3.14.0

on GitHub

🚀 Major Enhancements

API Server Improvements

• Increased file upload limit to 1GB - Resolves 413 errors for large document processing
• Comprehensive OpenAPI documentation with Swagger UI, ReDoc, and multiple interface options
• Type-safe API responses using TypedDict instead of generic dict[str, Any]
• Enhanced error handling with detailed debugging support via DEBUG environment variable

DPI Configuration System

• Automatic image size optimization prevents "Image too large" OCR failures
• Configurable DPI settings with intelligent auto-adjustment for large documents
• Performance vs quality trade-offs with customizable target DPI, min/max bounds
• Seamless integration across all OCR backends (Tesseract, EasyOCR, PaddleOCR)

Infrastructure & Testing

• Comprehensive test reorganization - Tests now organized in logical folders (core, features, integration)
• Enhanced CI pipeline with improved coverage combining and error handling
• New DPI integration tests ensuring robust image processing across all document types
• Quality assurance improvements with expanded regression testing

Documentation Updates

• Updated extraction configuration docs with DPI system examples and best practices
• Enhanced OCR configuration guide with performance optimization recommendations
• Complete API server documentation including OpenAPI endpoint references
• User-focused approach emphasizing practical usage over implementation details

📋 Technical Details

New Configuration Options

config = ExtractionConfig(
    target_dpi=150,              # Optimal balance of quality and performance
    max_image_dimension=25000,   # Prevent memory issues with large documents
    auto_adjust_dpi=True,        # Intelligent scaling for oversized images
    min_dpi=72,                  # Minimum readable resolution
    max_dpi=600                  # Maximum before diminishing returns
)

API Improvements

• 1GB file upload support via request_max_body_size configuration
• OpenAPI documentation available at multiple endpoints (/schema/swagger, /schema/redoc, etc.)
• Structured error responses with context for debugging

🔧 Breaking Changes

None - all changes are backward compatible.

🐛 Bug Fixes

• Fixed CI coverage combining issues with lcov command syntax
• Resolved pre-commit formatting conflicts
• Eliminated "Image too large" errors across all OCR backends

Full Changelog: v3.13.4...v3.14.0