kreuzberg 3.8.0__tar.gz → 3.8.1__tar.gz

{kreuzberg-3.8.0 → kreuzberg-3.8.1}/PKG-INFO

@@ -1,14 +1,16 @@
 Metadata-Version: 2.4
 Name: kreuzberg
-Version: 3.8.0
-Summary: A text extraction library supporting PDFs, images, office documents and more
+Version: 3.8.1
+Summary: Advanced document intelligence framework for extracting structured content from PDFs, images, and office documents
 Project-URL: homepage, https://github.com/Goldziher/kreuzberg
 Author-email: Na'aman Hirschfeld <nhirschfed@gmail.com>
 License: MIT
 License-File: LICENSE
-Keywords: document-processing,entity-extraction,image-to-text,keyword-extraction,named-entity-recognition,ner,ocr,pandoc,pdf-extraction,rag,spacy,table-extraction,tesseract,text-extraction,text-processing
+Keywords: automation,content-extraction,data-processing,document-analysis,document-intelligence,document-processing,entity-extraction,image-to-text,information-extraction,ocr,pdf-extraction,rag,structured-data,table-extraction,text-extraction
 Classifier: Development Status :: 5 - Production/Stable
 Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: Information Technology
+Classifier: Intended Audience :: Science/Research
 Classifier: License :: OSI Approved :: MIT License
 Classifier: Operating System :: OS Independent
 Classifier: Programming Language :: Python :: 3 :: Only
@@ -16,10 +18,13 @@ Classifier: Programming Language :: Python :: 3.10
 Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Database
+Classifier: Topic :: Multimedia :: Graphics :: Capture :: Scanners
+Classifier: Topic :: Office/Business :: Office Suites
 Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Scientific/Engineering :: Information Analysis
 Classifier: Topic :: Software Development :: Libraries :: Python Modules
 Classifier: Topic :: Text Processing :: General
-Classifier: Topic :: Utilities
 Classifier: Typing :: Typed
 Requires-Python: >=3.10
 Requires-Dist: anyio>=4.9.0
@@ -83,49 +88,31 @@ Description-Content-Type: text/markdown
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 [![Test Coverage](https://img.shields.io/badge/coverage-95%25-green)](https://github.com/Goldziher/kreuzberg)
 
-**High-performance Open Source Document Intelligence framework for Python.** Built by engineers for production workloads - extract text from any document with excellent performance and minimal complexity.
+**Advanced Document Intelligence for Modern Python Applications.** Transform PDFs, images, and office documents into structured data with production-grade performance. Built by engineers who understand that speed, reliability, and developer experience matter.
 
 📖 **[Complete Documentation](https://goldziher.github.io/kreuzberg/)**
 
 ## Why Choose Kreuzberg?
 
-### 🚀 Performance
+### ⚡ Proven Performance
 
-- [benchmarked as the fastest framework](https://goldziher.github.io/python-text-extraction-libs-benchmarks/) - 2-3x faster than the nearest alternatives
-- Minimal footprint: 71MB install vs 1GB+ for competitors
-- Lowest memory usage (~530MB average) optimized for production workloads
-- Edge and serverless ready - deploy anywhere without heavy dependencies
+[Benchmarked](https://goldziher.github.io/python-text-extraction-libs-benchmarks/) 6-126x faster than alternatives while using minimal resources. Process up to 14 files per second with 87MB install size and ~360MB memory usage. Optimized for production workloads and resource-constrained environments.
 
-### 🛠️ Engineering Quality
+### 🏗️ Production Engineering
 
-- Built by software engineers with modern Python best practices
-- 95%+ test coverage with comprehensive test suite
-- Thoroughly benchmarked and profiled for real-world performance
-- Only framework offering true async/await support alongside sync APIs
-- Robust error handling and detailed logging
+Comprehensive test coverage (95%+), robust error handling, and true async/await support. Built with modern Python practices for reliability in production environments.
 
-### 🎯 Developer Experience
+### 🔧 Developer Experience
 
-- Works out of the box with sane defaults, scales with your needs
-- Native MCP server for AI tool integration (Claude Desktop, Cursor)
-- Full type safety with excellent IDE support (completions)
-- Comprehensive documentation including full API reference
+Works immediately with smart defaults, scales as you grow. Native MCP integration for AI tools, full type safety, and clear documentation.
 
-### 🌍 Deployment Options
+### 🚀 Flexible Deployment
 
-- Docker images for all architectures (AMD64, ARM64)
-- Cloud native - AWS Lambda, Google Cloud Functions, Azure Functions
-- CPU-only processing - no GPU requirements, lower energy consumption
-- 100% local processing - no external API dependencies
-- Multiple deployment modes: CLI, REST API, MCP server
+Deploy on serverless platforms, containers, or traditional servers. Supports both CPU and GPU processing (via PaddleOCR and EasyOCR). No external API dependencies. Multiple deployment modes: CLI, REST API, MCP server.
 
-### 🎯 Complete Solution
+### 📄 Comprehensive Format Support
 
-- Universal format support: PDFs, images, Office docs, HTML, spreadsheets, presentations
-- Multiple OCR engines: Tesseract, EasyOCR, PaddleOCR with intelligent fallbacks
-- Advanced features: Table extraction, metadata extraction, content chunking for RAG
-- Production tools: REST API, CLI tools, batch processing, custom extractors
-- Fully extensible: Add your own extractors
+Extract from PDFs, images, Office documents, HTML, spreadsheets, and presentations. Multiple OCR engines with intelligent fallbacks, table extraction, and content preparation for RAG workflows.
 
 ## Quick Start
 
@@ -161,7 +148,7 @@ import asyncio
 from kreuzberg import extract_file
 
 async def main():
-    # Extract from any document type
+    # Extract content from files
     result = await extract_file("document.pdf")
     print(result.content)
     print(result.metadata)
@@ -275,23 +262,23 @@ kreuzberg extract *.pdf --output-dir ./extracted/
 
 ## 📊 Performance Comparison
 
-[Comprehensive benchmarks](https://goldziher.github.io/python-text-extraction-libs-benchmarks/) across 94 real-world documents • [View source](https://github.com/Goldziher/python-text-extraction-libs-benchmarks):
+[Comprehensive benchmarks](https://goldziher.github.io/python-text-extraction-libs-benchmarks/) across ~100 real-world documents • [View source](https://github.com/Goldziher/python-text-extraction-libs-benchmarks) • [**Detailed Analysis**](https://goldziher.github.io/kreuzberg/performance-analysis/):
 
-| Framework     | Speed       | Memory | Install Size | Dependencies | Success Rate |
-| ------------- | ----------- | ------ | ------------ | ------------ | ------------ |
-| **Kreuzberg** | 35+ files/s | 530MB  | 71MB         | 20           | High         |
-| Unstructured  | ~12 files/s | ~1GB   | 146MB        | 54           | 88%+         |
-| MarkItDown    | ~15 files/s | ~1.5GB | 251MB        | 25           | 80%\*        |
-| Docling       | ~1 file/min | ~5GB   | 1,032MB      | 88           | 45%\*        |
+| Framework     | Speed        | Memory | Install Size | Dependencies | Success Rate |
+| ------------- | ------------ | ------ | ------------ | ------------ | ------------ |
+| **Kreuzberg** | 14.4 files/s | 360MB  | 87MB         | 43           | 100%         |
+| Unstructured  | ~12 files/s  | ~1GB   | 146MB        | 54           | 88%+         |
+| MarkItDown    | ~15 files/s  | ~1.5GB | 251MB        | 25           | 80%\*        |
+| Docling       | ~1 file/min  | ~5GB   | 1,032MB      | 88           | 45%\*        |
 
 \*_Performance varies significantly with document complexity and size_
 
 **Key strengths:**
 
-- 2-3x faster processing than comparable frameworks
+- 6-126x faster processing than comparable frameworks
 - Smallest installation footprint and memory usage
 - Only framework with built-in async/await support
-- CPU-only processing - no GPU dependencies
+- Supports both CPU and GPU processing
 - Built by software engineers for production reliability
 
 > **Benchmark details**: Tests include PDFs, Word docs, HTML, images, and spreadsheets in multiple languages (English, Hebrew, German, Chinese, Japanese, Korean) on standardized hardware.
@@ -302,6 +289,7 @@ kreuzberg extract *.pdf --output-dir ./extracted/
 
 - [Installation Guide](https://goldziher.github.io/kreuzberg/getting-started/installation/) - Setup and dependencies
 - [User Guide](https://goldziher.github.io/kreuzberg/user-guide/) - Comprehensive usage guide
+- [Performance Analysis](https://goldziher.github.io/kreuzberg/performance-analysis/) - Detailed benchmark results
 - [API Reference](https://goldziher.github.io/kreuzberg/api-reference/) - Complete API documentation
 - [Docker Guide](https://goldziher.github.io/kreuzberg/user-guide/docker/) - Container deployment
 - [REST API](https://goldziher.github.io/kreuzberg/user-guide/api-server/) - HTTP endpoints

{kreuzberg-3.8.0 → kreuzberg-3.8.1}/README.md

@@ -6,49 +6,31 @@
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 [![Test Coverage](https://img.shields.io/badge/coverage-95%25-green)](https://github.com/Goldziher/kreuzberg)
 
-**High-performance Open Source Document Intelligence framework for Python.** Built by engineers for production workloads - extract text from any document with excellent performance and minimal complexity.
+**Advanced Document Intelligence for Modern Python Applications.** Transform PDFs, images, and office documents into structured data with production-grade performance. Built by engineers who understand that speed, reliability, and developer experience matter.
 
 📖 **[Complete Documentation](https://goldziher.github.io/kreuzberg/)**
 
 ## Why Choose Kreuzberg?
 
-### 🚀 Performance
+### ⚡ Proven Performance
 
-- [benchmarked as the fastest framework](https://goldziher.github.io/python-text-extraction-libs-benchmarks/) - 2-3x faster than the nearest alternatives
-- Minimal footprint: 71MB install vs 1GB+ for competitors
-- Lowest memory usage (~530MB average) optimized for production workloads
-- Edge and serverless ready - deploy anywhere without heavy dependencies
+[Benchmarked](https://goldziher.github.io/python-text-extraction-libs-benchmarks/) 6-126x faster than alternatives while using minimal resources. Process up to 14 files per second with 87MB install size and ~360MB memory usage. Optimized for production workloads and resource-constrained environments.
 
-### 🛠️ Engineering Quality
+### 🏗️ Production Engineering
 
-- Built by software engineers with modern Python best practices
-- 95%+ test coverage with comprehensive test suite
-- Thoroughly benchmarked and profiled for real-world performance
-- Only framework offering true async/await support alongside sync APIs
-- Robust error handling and detailed logging
+Comprehensive test coverage (95%+), robust error handling, and true async/await support. Built with modern Python practices for reliability in production environments.
 
-### 🎯 Developer Experience
+### 🔧 Developer Experience
 
-- Works out of the box with sane defaults, scales with your needs
-- Native MCP server for AI tool integration (Claude Desktop, Cursor)
-- Full type safety with excellent IDE support (completions)
-- Comprehensive documentation including full API reference
+Works immediately with smart defaults, scales as you grow. Native MCP integration for AI tools, full type safety, and clear documentation.
 
-### 🌍 Deployment Options
+### 🚀 Flexible Deployment
 
-- Docker images for all architectures (AMD64, ARM64)
-- Cloud native - AWS Lambda, Google Cloud Functions, Azure Functions
-- CPU-only processing - no GPU requirements, lower energy consumption
-- 100% local processing - no external API dependencies
-- Multiple deployment modes: CLI, REST API, MCP server
+Deploy on serverless platforms, containers, or traditional servers. Supports both CPU and GPU processing (via PaddleOCR and EasyOCR). No external API dependencies. Multiple deployment modes: CLI, REST API, MCP server.
 
-### 🎯 Complete Solution
+### 📄 Comprehensive Format Support
 
-- Universal format support: PDFs, images, Office docs, HTML, spreadsheets, presentations
-- Multiple OCR engines: Tesseract, EasyOCR, PaddleOCR with intelligent fallbacks
-- Advanced features: Table extraction, metadata extraction, content chunking for RAG
-- Production tools: REST API, CLI tools, batch processing, custom extractors
-- Fully extensible: Add your own extractors
+Extract from PDFs, images, Office documents, HTML, spreadsheets, and presentations. Multiple OCR engines with intelligent fallbacks, table extraction, and content preparation for RAG workflows.
 
 ## Quick Start
 
@@ -84,7 +66,7 @@ import asyncio
 from kreuzberg import extract_file
 
 async def main():
-    # Extract from any document type
+    # Extract content from files
     result = await extract_file("document.pdf")
     print(result.content)
     print(result.metadata)
@@ -198,23 +180,23 @@ kreuzberg extract *.pdf --output-dir ./extracted/
 
 ## 📊 Performance Comparison
 
-[Comprehensive benchmarks](https://goldziher.github.io/python-text-extraction-libs-benchmarks/) across 94 real-world documents • [View source](https://github.com/Goldziher/python-text-extraction-libs-benchmarks):
+[Comprehensive benchmarks](https://goldziher.github.io/python-text-extraction-libs-benchmarks/) across ~100 real-world documents • [View source](https://github.com/Goldziher/python-text-extraction-libs-benchmarks) • [**Detailed Analysis**](https://goldziher.github.io/kreuzberg/performance-analysis/):
 
-| Framework     | Speed       | Memory | Install Size | Dependencies | Success Rate |
-| ------------- | ----------- | ------ | ------------ | ------------ | ------------ |
-| **Kreuzberg** | 35+ files/s | 530MB  | 71MB         | 20           | High         |
-| Unstructured  | ~12 files/s | ~1GB   | 146MB        | 54           | 88%+         |
-| MarkItDown    | ~15 files/s | ~1.5GB | 251MB        | 25           | 80%\*        |
-| Docling       | ~1 file/min | ~5GB   | 1,032MB      | 88           | 45%\*        |
+| Framework     | Speed        | Memory | Install Size | Dependencies | Success Rate |
+| ------------- | ------------ | ------ | ------------ | ------------ | ------------ |
+| **Kreuzberg** | 14.4 files/s | 360MB  | 87MB         | 43           | 100%         |
+| Unstructured  | ~12 files/s  | ~1GB   | 146MB        | 54           | 88%+         |
+| MarkItDown    | ~15 files/s  | ~1.5GB | 251MB        | 25           | 80%\*        |
+| Docling       | ~1 file/min  | ~5GB   | 1,032MB      | 88           | 45%\*        |
 
 \*_Performance varies significantly with document complexity and size_
 
 **Key strengths:**
 
-- 2-3x faster processing than comparable frameworks
+- 6-126x faster processing than comparable frameworks
 - Smallest installation footprint and memory usage
 - Only framework with built-in async/await support
-- CPU-only processing - no GPU dependencies
+- Supports both CPU and GPU processing
 - Built by software engineers for production reliability
 
 > **Benchmark details**: Tests include PDFs, Word docs, HTML, images, and spreadsheets in multiple languages (English, Hebrew, German, Chinese, Japanese, Korean) on standardized hardware.
@@ -225,6 +207,7 @@ kreuzberg extract *.pdf --output-dir ./extracted/
 
 - [Installation Guide](https://goldziher.github.io/kreuzberg/getting-started/installation/) - Setup and dependencies
 - [User Guide](https://goldziher.github.io/kreuzberg/user-guide/) - Comprehensive usage guide
+- [Performance Analysis](https://goldziher.github.io/kreuzberg/performance-analysis/) - Detailed benchmark results
 - [API Reference](https://goldziher.github.io/kreuzberg/api-reference/) - Complete API documentation
 - [Docker Guide](https://goldziher.github.io/kreuzberg/user-guide/docker/) - Container deployment
 - [REST API](https://goldziher.github.io/kreuzberg/user-guide/api-server/) - HTTP endpoints

{kreuzberg-3.8.0 → kreuzberg-3.8.1}/docs/index.md

@@ -1,23 +1,19 @@
 # Kreuzberg
 
-Kreuzberg is a complete Open Source Document Intelligence framework. Its Built by engineers for production workloads -
-its not a data science / research orientated tool, but rather a pragmatic swiss-army knife that is meant to deliver.
-Yes, Python, when coupled with robust technologies such as `pdfium`, `tesseract` and `pandoc` can do quite a lot.
-Kreuzberg was also created (primarily) in Kreuzberg - the famous and beautiful neighborhood of Berlin.
+Kreuzberg is an advanced open source document intelligence framework built for production workloads. Designed by engineers for reliability and performance, it transforms PDFs, images, and office documents into structured data with minimal complexity.
+
+Built on proven technologies including PDFium, Tesseract, and Pandoc, Kreuzberg delivers enterprise-grade document processing capabilities while maintaining simplicity and speed.
 
 ## Why Kreuzberg?
 
-At the danger of over-selling, there are actually quite a lot of reasons why use Kreuzberg. You can read them below.
-BUT - this is not necessarily a mutually exclusive solution. For example.
-many text extraction pipelines can integrate a library such as Kreuzberg with some kind of heuristics on when to use it
-and when use something else.
+Kreuzberg addresses real production needs with measurable benefits. While not exclusively a complete solution, it integrates well with existing pipelines and can be deployed alongside other tools based on specific requirements.
 
 ### 🚀 Performance
 
-- [benchmarked as the fastest framework](https://goldziher.github.io/python-text-extraction-libs-benchmarks/) - 2-3x
-    faster than the nearest alternatives
-- Minimal footprint: 71MB install vs 1GB+ for competitors
-- Lowest memory usage (~530MB average) optimized for production workloads
+- [benchmarked as the fastest framework](https://goldziher.github.io/python-text-extraction-libs-benchmarks/) - 6-126x
+    faster than competitors
+- Minimal footprint: 87MB install vs 1GB+ for competitors
+- Lowest memory usage (~360MB average) optimized for production workloads
 - Edge and serverless ready - deploy anywhere without heavy dependencies
 
 ### 🛠️ Engineering Quality
@@ -39,13 +35,13 @@ and when use something else.
 
 - Docker images for all architectures (AMD64, ARM64)
 - Cloud native - AWS Lambda, Google Cloud Functions, Azure Functions
-- CPU-only processing - no GPU requirements, lower energy consumption
-- 100% local processing - no external API dependencies
+- Supports both CPU and GPU processing (PaddleOCR, EasyOCR)
+- Local processing - no external API dependencies
 - Multiple deployment modes: CLI, REST API, MCP server
 
 ### 🎯 Complete Solution
 
-- Universal format support: PDFs, images, Office docs, HTML, spreadsheets, presentations
+- Comprehensive format support: PDFs, images, Office docs, HTML, spreadsheets, presentations
 - Multiple OCR engines: Tesseract, EasyOCR, PaddleOCR with intelligent fallbacks
 - Advanced features: Table extraction, metadata extraction, content chunking for RAG
 - Production tools: REST API, CLI tools, batch processing, custom extractors

kreuzberg-3.8.1/docs/performance-analysis.md

@@ -0,0 +1,140 @@
+# Performance Analysis
+
+## Overview
+
+This page presents comprehensive benchmark results comparing Kreuzberg against other text extraction frameworks. All data is derived from rigorous testing across ~100 real-world documents using standardized methodology.
+
+> **Benchmark Methodology**: Results based on the [python-text-extraction-libraries-benchmarks-2025](https://github.com/Goldziher/python-text-extraction-libraries-benchmarks-2025) project with comprehensive testing across multiple document types and sizes.
+
+## Executive Summary
+
+Kreuzberg demonstrates exceptional performance across all key metrics:
+
+- **Speed**: 6-126x faster than competitors
+- **Memory**: 2-4x lower usage
+- **Installation**: 2-68x smaller footprint
+- **Reliability**: Perfect 100% success rate
+
+## Detailed Performance Metrics
+
+### Processing Speed
+
+#### By File Size Category
+
+| Category              | Kreuzberg Sync | Kreuzberg Async | Best Competitor | Advantage   |
+| --------------------- | -------------- | --------------- | --------------- | ----------- |
+| **Tiny (\<100KB)**    | 31.6 files/sec | 23.6 files/sec  | 4.8 files/sec   | 6.6x faster |
+| **Small (100KB-1MB)** | 9.0 files/sec  | 10.1 files/sec  | 3.6 files/sec   | 2.8x faster |
+| **Medium (1-10MB)**   | 2.6 files/sec  | 3.2 files/sec   | 0.065 files/sec | 49x faster  |
+
+#### Processing Time Comparison
+
+| Framework           | Tiny Files (s) | Small Files (s) | Medium Files (s) |
+| ------------------- | -------------- | --------------- | ---------------- |
+| **Kreuzberg Sync**  | 0.032          | 0.111           | 0.388            |
+| **Kreuzberg Async** | 0.042          | 0.099           | 0.315            |
+| Extractous          | 0.316          | 0.281           | 15.38            |
+| Unstructured        | 0.210          | 1.123           | -                |
+| Docling             | 3.956          | 14.47           | -                |
+
+### Memory Usage
+
+| Framework           | Average Memory (MB) | vs Kreuzberg |
+| ------------------- | ------------------- | ------------ |
+| **Kreuzberg Sync**  | 360                 | Baseline     |
+| **Kreuzberg Async** | 396                 | +10%         |
+| Extractous          | 513                 | +43%         |
+| Unstructured        | 1,389               | +286%        |
+| Docling             | 1,838               | +411%        |
+
+### Installation Size
+
+| Framework     | Size (MB) | Packages | vs Kreuzberg |
+| ------------- | --------- | -------- | ------------ |
+| **Kreuzberg** | 87        | 43       | Baseline     |
+| Unstructured  | 176       | 54       | 2.0x larger  |
+| MarkItDown    | 208       | 25       | 2.4x larger  |
+| Docling       | 5,900     | 103      | 67.8x larger |
+
+### Success Rate & Reliability
+
+| Framework     | Tiny Files | Small Files | Medium Files | Overall  |
+| ------------- | ---------- | ----------- | ------------ | -------- |
+| **Kreuzberg** | 100%       | 100%        | 100%         | **100%** |
+| Extractous    | 100%       | 95.8%       | 100%         | 98.6%    |
+| Unstructured  | 100%       | 100%        | -            | 100%     |
+| Docling       | 100%       | 96.3%       | -            | 98.2%    |
+
+### Content Extraction Quality
+
+#### Characters Extracted (Average)
+
+| Framework     | Tiny Files | Small Files | Medium Files |
+| ------------- | ---------- | ----------- | ------------ |
+| **Kreuzberg** | 6,950      | 173,505     | 500,643      |
+| Extractous    | 6,894      | 106,641     | 251,612      |
+| Unstructured  | 3,842      | 70,396      | -            |
+| Docling       | 3,316      | 59,129      | -            |
+
+## Performance Insights
+
+### Speed Advantages
+
+1. **Optimized Processing Pipeline**: Efficient async/await implementation
+1. **Smart Resource Management**: Minimal overhead operations
+1. **Native Libraries**: Built on high-performance C libraries (PDFium, Tesseract)
+
+### Memory Efficiency
+
+1. **Lean Architecture**: Minimal memory footprint during processing
+1. **Resource Cleanup**: Proper resource disposal and garbage collection
+1. **Streaming Processing**: Process large files without loading entirely into memory
+
+### Installation Benefits
+
+1. **Minimal Dependencies**: Only essential packages included
+1. **No Heavy ML Models**: CPU-focused processing without large model files
+1. **Efficient Packaging**: Optimized distribution with selective dependencies
+
+## Production Implications
+
+### Cost Savings
+
+- **Infrastructure**: 2-4x lower memory requirements reduce server costs
+- **Storage**: 2-68x smaller installation saves disk space
+- **Processing**: 6-126x faster execution reduces compute time
+
+### Operational Benefits
+
+- **Deployment Speed**: Faster installations and updates
+- **Resource Planning**: Predictable memory and CPU usage
+- **Scaling**: Efficient resource utilization enables higher throughput
+
+### Developer Experience
+
+- **Quick Setup**: Minimal installation time and complexity
+- **Reliable Performance**: Consistent results across document types
+- **Production Ready**: Battle-tested performance characteristics
+
+## Test Environment
+
+**Hardware**: Linux CI runners
+**Python Version**: 3.13
+**Document Corpus**: ~100 real-world documents tested across multiple frameworks
+**Test Date**: July 13, 2025
+**Methodology**: [Full methodology available](https://github.com/Goldziher/python-text-extraction-libraries-benchmarks-2025)
+
+## Framework Comparison Matrix
+
+| Metric              | Kreuzberg | Extractous | Unstructured | Docling |
+| ------------------- | --------- | ---------- | ------------ | ------- |
+| **Speed**           | ★★★★★     | ★★☆☆☆      | ★★☆☆☆        | ★☆☆☆☆   |
+| **Memory**          | ★★★★★     | ★★★★☆      | ★★☆☆☆        | ★☆☆☆☆   |
+| **Installation**    | ★★★★★     | -          | ★★★☆☆        | ★☆☆☆☆   |
+| **Reliability**     | ★★★★★     | ★★★★☆      | ★★★★★        | ★★★★☆   |
+| **Content Quality** | ★★★★★     | ★★★☆☆      | ★★★☆☆        | ★★☆☆☆   |
+| **Overall**         | ★★★★★     | ★★★☆☆      | ★★★☆☆        | ★★☆☆☆   |
+
+______________________________________________________________________
+
+*Performance data is based on comprehensive benchmarking across real-world document corpus. Results may vary based on specific use cases and hardware configurations.*

{kreuzberg-3.8.0 → kreuzberg-3.8.1}/kreuzberg/_entity_extraction.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import os
 import re
 from dataclasses import dataclass
 from functools import lru_cache
@@ -181,8 +182,6 @@ def _load_spacy_model(model_name: str, spacy_config: SpacyEntityExtractionConfig
         import spacy
 
         if spacy_config.model_cache_dir:
-            import os
-
             os.environ["SPACY_DATA"] = str(spacy_config.model_cache_dir)
 
         nlp = spacy.load(model_name)

{kreuzberg-3.8.0 → kreuzberg-3.8.1}/kreuzberg/_extractors/_base.py

@@ -3,10 +3,12 @@ from __future__ import annotations
 from abc import ABC, abstractmethod
 from typing import TYPE_CHECKING, ClassVar
 
+from kreuzberg._types import ExtractionResult, normalize_metadata
+from kreuzberg._utils._quality import calculate_quality_score, clean_extracted_text
+
 if TYPE_CHECKING:
     from pathlib import Path
 
-    from kreuzberg import ExtractionResult
     from kreuzberg._types import ExtractionConfig
 
 
@@ -104,8 +106,6 @@ class Extractor(ABC):
         if not self.config.enable_quality_processing:
             return result
 
-        from kreuzberg._utils._quality import calculate_quality_score, clean_extracted_text
-
         if not result.content:
             return result
 
@@ -120,8 +120,6 @@ class Extractor(ABC):
         enhanced_metadata["quality_score"] = quality_score
 
         # Return enhanced result
-        from kreuzberg._types import ExtractionResult, normalize_metadata
-
         return ExtractionResult(
             content=cleaned_content,
             mime_type=result.mime_type,

{kreuzberg-3.8.0 → kreuzberg-3.8.1}/kreuzberg/_extractors/_image.py

@@ -11,13 +11,17 @@ from anyio import Path as AsyncPath
 from kreuzberg._extractors._base import Extractor
 from kreuzberg._mime_types import IMAGE_MIME_TYPES
 from kreuzberg._ocr import get_ocr_backend
-from kreuzberg._types import ExtractionResult
+from kreuzberg._ocr._easyocr import EasyOCRConfig
+from kreuzberg._ocr._paddleocr import PaddleOCRConfig
+from kreuzberg._ocr._tesseract import TesseractConfig
 from kreuzberg._utils._tmp import create_temp_file
 from kreuzberg.exceptions import ValidationError
 
 if TYPE_CHECKING:  # pragma: no cover
     from collections.abc import Mapping
 
+    from kreuzberg._types import ExtractionResult
+
 
 class ImageExtractor(Extractor):
     SUPPORTED_MIME_TYPES: ClassVar[set[str]] = IMAGE_MIME_TYPES
@@ -78,44 +82,26 @@ class ImageExtractor(Extractor):
         if self.config.ocr_backend is None:
             raise ValidationError("ocr_backend is None, cannot perform OCR")
 
-        if self.config.ocr_backend == "tesseract":
-            from kreuzberg._ocr._sync import process_batch_images_sync
-            from kreuzberg._ocr._tesseract import TesseractConfig
-
-            if isinstance(self.config.ocr_config, TesseractConfig):
-                config = self.config.ocr_config
-            else:
-                config = TesseractConfig()
-
-            results = process_batch_images_sync([str(path)], config, backend="tesseract")
-            if results:
-                result = results[0]
-                return self._apply_quality_processing(result)
-            return ExtractionResult(content="", mime_type="text/plain", metadata={}, chunks=[])
-
-        if self.config.ocr_backend == "paddleocr":
-            from kreuzberg._ocr._paddleocr import PaddleOCRConfig
-            from kreuzberg._ocr._sync import process_image_paddleocr_sync as paddle_process
+        backend = get_ocr_backend(self.config.ocr_backend)
 
+        if self.config.ocr_backend == "tesseract":
+            config = (
+                self.config.ocr_config if isinstance(self.config.ocr_config, TesseractConfig) else TesseractConfig()
+            )
+            result = backend.process_file_sync(path, **config.__dict__)
+        elif self.config.ocr_backend == "paddleocr":
             paddle_config = (
                 self.config.ocr_config if isinstance(self.config.ocr_config, PaddleOCRConfig) else PaddleOCRConfig()
             )
-
-            result = paddle_process(path, paddle_config)
-            return self._apply_quality_processing(result)
-
-        if self.config.ocr_backend == "easyocr":
-            from kreuzberg._ocr._easyocr import EasyOCRConfig
-            from kreuzberg._ocr._sync import process_image_easyocr_sync as easy_process
-
+            result = backend.process_file_sync(path, **paddle_config.__dict__)
+        elif self.config.ocr_backend == "easyocr":
             easy_config = (
                 self.config.ocr_config if isinstance(self.config.ocr_config, EasyOCRConfig) else EasyOCRConfig()
             )
-
-            result = easy_process(path, easy_config)
-            return self._apply_quality_processing(result)
-
-        raise NotImplementedError(f"Sync OCR not implemented for {self.config.ocr_backend}")
+            result = backend.process_file_sync(path, **easy_config.__dict__)
+        else:
+            raise NotImplementedError(f"Sync OCR not implemented for {self.config.ocr_backend}")
+        return self._apply_quality_processing(result)
 
     def _get_extension_from_mime_type(self, mime_type: str) -> str:
         if mime_type in self.IMAGE_MIME_TYPE_EXT_MAP:

{kreuzberg-3.8.0 → kreuzberg-3.8.1}/kreuzberg/_extractors/_pandoc.py

@@ -1,8 +1,11 @@
 from __future__ import annotations
 
 import contextlib
+import os
 import re
+import subprocess
 import sys
+import tempfile
 from json import JSONDecodeError, loads
 from pathlib import Path
 from typing import TYPE_CHECKING, Any, ClassVar, Final, Literal, cast
@@ -203,10 +206,6 @@ class PandocExtractor(Extractor):
         Returns:
             ExtractionResult with the extracted text and metadata.
         """
-        import os
-        import tempfile
-        from pathlib import Path
-
         extension = self._get_pandoc_type_from_mime_type(self.mime_type)
         fd, temp_path = tempfile.mkstemp(suffix=f".{extension}")
 
@@ -579,8 +578,6 @@ class PandocExtractor(Extractor):
 
     def _validate_pandoc_version_sync(self) -> None:
         """Synchronous version of _validate_pandoc_version."""
-        import subprocess
-
         try:
             if self._checked_version:
                 return
@@ -625,10 +622,6 @@ class PandocExtractor(Extractor):
 
     def _extract_metadata_sync(self, path: Path) -> Metadata:
         """Synchronous version of _handle_extract_metadata."""
-        import os
-        import subprocess
-        import tempfile
-
         pandoc_type = self._get_pandoc_type_from_mime_type(self.mime_type)
         fd, metadata_file = tempfile.mkstemp(suffix=".json")
         os.close(fd)
@@ -663,10 +656,6 @@ class PandocExtractor(Extractor):
 
     def _extract_file_sync(self, path: Path) -> str:
         """Synchronous version of _handle_extract_file."""
-        import os
-        import subprocess
-        import tempfile
-
         pandoc_type = self._get_pandoc_type_from_mime_type(self.mime_type)
         fd, output_path = tempfile.mkstemp(suffix=".md")
         os.close(fd)