A polyglot document intelligence framework with a Rust core. Extract text, metadata, and structured information from PDFs, Office documents, images, and 56 formats. Available for Rust, Python, TypeScript/Node.js, Ruby, Go, Java, and C#—or use via CLI, REST API, or MCP server.

🚀 Version 4.0.0 Release Candidate This is a pre-release version. We invite you to test the library and report any issues you encounter. Help us make the stable release better!

⚠️ BREAKING CHANGE in RC.11

Embeddings now require separate ONNX Runtime installation. This reduces package sizes and enables Windows MSVC support.

• Action required: Install ONNX Runtime if you use embeddings
• No action needed: If you don't use embeddings
• Installation Guide

• Truly polyglot – Native bindings for Rust, Python, TypeScript/Node.js, Ruby, Go, Java, C#
• Production-ready – Battle-tested with comprehensive error handling and validation
• 56 formats – PDF, Office documents, images, HTML, XML, emails, archives, and more
• OCR built-in – Multiple backends (Tesseract, EasyOCR, PaddleOCR) with table extraction support
• Flexible deployment – Use as library, CLI tool, REST API server, or MCP server
• Memory efficient – Streaming parsers with constant memory usage for multi-GB files

📖 Complete Documentation • 🚀 Installation Guides

Don't want to manage Rust infrastructure? Kreuzberg Cloud is a managed document extraction API launching at the beginning of 2026.

• Hosted REST API with async jobs and webhooks
• Built-in chunking and embeddings for RAG pipelines
• Premium OCR backends for 95%+ accuracy
• No infrastructure to maintain

Each language binding provides comprehensive documentation with examples and best practices. Choose your platform to get started:

• @kreuzberg/node (Recommended for Node.js/Bun) – Native NAPI-RS bindings, fastest performance, direct system calls
• @kreuzberg/wasm (Browser/Workers/Deno) – Pure WebAssembly, no native dependencies, cross-platform consistency

TypeScript Decision Matrix:

• Python – Installation, basic usage, async/sync APIs
• Ruby – Installation, basic usage, configuration
• Go – Installation, native library setup, sync/async extraction + batch APIs Note: Windows builds use MinGW and don't support embeddings (ONNX Runtime requires MSVC)
• Java – Installation, FFM API usage, Maven/Gradle setup
• C# – Installation, P/Invoke usage, NuGet package
• Rust – Crate usage, features, async/sync APIs
• CLI – Command-line usage, batch processing, options

To use embeddings functionality:

1. Install ONNX Runtime:

   • Linux: apt install libonnxruntime
   • macOS: brew install onnxruntime
   • Windows: scoop install onnxruntime or winget install onnxruntime

2. Use embeddings in your code - see Embeddings Guide

Note: All other Kreuzberg features work without ONNX Runtime.

The Rust crate offers flexible PDFium linking strategies for different deployment scenarios. Language bindings (Python, TypeScript, Ruby, Java, Go, C#) always bundle PDFium automatically—no configuration needed.

Example Cargo.toml configurations:

# Default (dynamic linking)
[dependencies]
kreuzberg = "4.0"

# Static linking
[dependencies]
kreuzberg = { version = "4.0", features = ["pdf-static"] }

# Bundled in binary
[dependencies]
kreuzberg = { version = "4.0", features = ["pdf-bundled"] }

# System library
[dependencies]
kreuzberg = { version = "4.0", features = ["pdf-system"] }

Complete PDFium Linking Guide → for detailed configuration, trade-offs, and troubleshooting.

All image formats support OCR: .jpg, .jpeg, .png, .tiff, .tif, .bmp, .gif, .webp, .jp2

LaTeX (.tex), BibTeX (.bib), Jupyter (.ipynb), reStructuredText (.rst), Org Mode (.org), Markdown (.md)

Complete Format Documentation

Multiple OCR backends (Tesseract, EasyOCR, PaddleOCR) with intelligent table detection and reconstruction. Extract structured data from scanned documents and images with configurable accuracy thresholds.

OCR Backend Documentation →

Process multiple documents concurrently with configurable parallelism. Optimize throughput for large-scale document processing workloads with automatic resource management.

Batch Processing Guide →

Handle encrypted PDFs with single or multiple password attempts. Supports both RC4 and AES encryption with automatic fallback strategies.

PDF Configuration →

Automatic language detection in extracted text using fast-langdetect. Configure confidence thresholds and access per-language statistics.

Language Detection Guide →

Extract comprehensive metadata from all supported formats: authors, titles, creation dates, page counts, EXIF data, and format-specific properties.

Metadata Guide →

Kreuzberg offers two JavaScript/TypeScript options with different performance characteristics:

Use @kreuzberg/node when:

• Running on Node.js or Bun backend servers
• Performance is critical (2-3x faster than WASM)
• You have or can install native build toolchain
• Processing high document volumes

Use @kreuzberg/wasm when:

• Running in browser or web workers
• Deploying to serverless edge (Cloudflare, Vercel)
• Using Deno or similar runtimes
• Need absolute zero native dependencies
• Cross-platform consistency is important
• Trade-off: ~20-40% slower but still good performance

Production-ready API server with OpenAPI documentation, health checks, and telemetry support. Deploy standalone or in containers with automatic format detection and streaming support.

API Server Documentation →

Model Context Protocol server for Claude and other AI assistants. Enables AI agents to extract and process documents directly with full configuration support.

MCP Server Documentation →

Official Docker images available in multiple variants:

• Core (~1.0-1.3GB): Tesseract OCR, modern Office formats
• Full (~1.5-2.1GB): Adds LibreOffice for legacy Office formats (.doc, .ppt)

All images support API server, CLI, and MCP server modes with automatic platform detection for linux/amd64 and linux/arm64.

Docker Deployment Guide →

When to use Kreuzberg:

• ✅ Need high throughput (thousands of documents)
• ✅ Memory-constrained environments
• ✅ Non-Python ecosystems (Ruby, TypeScript, Java, Go)
• ✅ RAG pipelines (built-in chunking + embeddings)
• ✅ Self-hosted or air-gapped deployments
• ✅ Multi-GB files requiring streaming

When to consider alternatives:

• LlamaParse: If you need best-in-class table extraction and only process PDFs (requires internet, paid)
• docling: If you're Python-only and don't need extreme performance
• unstructured: If you need extensive pre-built integrations with vector databases

Kreuzberg is built with a Rust core for efficient document extraction and processing.

• Rust core – Native code for text extraction and processing
• Async throughout – Asynchronous processing with Tokio runtime
• Memory efficient – Streaming parsers for large files
• Parallel batch processing – Configurable concurrency for multiple documents
• Zero-copy operations – Efficient data handling where possible

• Installation Guide – Setup and dependencies
• User Guide – Comprehensive usage guide
• API Reference – Complete API documentation
• Format Support – Supported file formats
• OCR Backends – OCR engine setup
• CLI Guide – Command-line usage
• Migration Guide – Upgrading from v3

Contributions are welcome! See CONTRIBUTING.md for guidelines.

MIT License - see LICENSE for details. Kreuzberg’s open-source is released under the MIT license, one of the most permissive licenses available ❤️ This means you can use it freely in both commercial and closed-source products with no obligations, no viral effects, and no licensing restrictions.

Unlike AGPL-licensed PDF engines like MuPDF (which require you to open-source your entire codebase unless you buy a commercial license), MIT is safe for enterprise adoption and creates zero legal friction.