Kreuzberg

Extract text, metadata, and code intelligence from 90+ file formats and 300+ programming languages at native speeds without needing a GPU.

• Code intelligence – Extract functions, classes, imports, symbols, and docstrings from 300+ programming languages via tree-sitter. Results in ExtractionResult.code_intelligence with semantic chunking
• Extensible architecture – Plugin system for custom OCR backends, validators, post-processors, document extractors, and renderers
• Polyglot – Native bindings for Rust, Python, TypeScript/Node.js, Ruby, Go, Java, Kotlin, C#, PHP, Elixir, R, Dart, Swift, Zig, and C
• 90+ file formats – PDF, Office documents, images, HTML, XML, emails, archives, academic formats across 8 categories
• LLM intelligence – VLM OCR (GPT-4o, Claude, Gemini, Ollama), structured JSON extraction with schema constraints, and provider-hosted embeddings via 143 LLM providers (including local engines: Ollama, LM Studio, vLLM, llama.cpp) through liter-llm
• OCR support – Tesseract (all bindings, including Tesseract-WASM for browsers), PaddleOCR (all native bindings), EasyOCR (Python), VLM OCR (143 vision model providers including local engines), extensible via plugin API
• High performance – Rust core with pure-Rust PDF, SIMD optimizations and full parallelism
• Flexible deployment – Use as library, CLI tool, REST API server, or MCP server
• TOON wire format – Token-efficient serialization for LLM/RAG pipelines, ~30-50% fewer tokens than JSON
• GFM-quality output – Comrak-based rendering with proper fenced code blocks, table nodes, bracket escaping, and cross-format parity (Markdown, HTML, Djot, Plain)
• HTML passthrough – HTML-to-Markdown conversion uses html-to-markdown output directly, bypassing lossy intermediate round-trips
• Memory efficient – Streaming parsers for multi-GB files

Complete Documentation | Live Demo | Installation Guides

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

Scripting Languages:

• Python – PyPI package, async/sync APIs, OCR backends (Tesseract, PaddleOCR, EasyOCR)
• Ruby – RubyGems package, idiomatic Ruby API, native bindings
• PHP – Composer package, modern PHP 8.2+ support, type-safe API, async extraction
• Elixir – Hex package, OTP integration, concurrent processing
• R – r-universe package, idiomatic R API, extendr bindings
• Dart / Flutter – pub.dev package, flutter_rust_bridge runtime, native bindings for macOS/iOS/Android/Linux/Windows

JavaScript/TypeScript:

• @kreuzberg/node – Native NAPI-RS bindings for Node.js/Bun, fastest performance
• @kreuzberg/wasm – WebAssembly for browsers/Deno/Cloudflare Workers, comprehensive format and OCR support (PDF, Excel, archives, all office formats, real Tesseract via the WASI build) — only ORT-dependent features (paddle-ocr, layout detection, embeddings, auto-rotate) and server modes (api/mcp/cli) are excluded

Compiled Languages:

• Go – Go module with FFI bindings, context-aware async
• Java – Maven Central, Foreign Function & Memory API
• Kotlin – Maven Central, Kotlin/JVM with idiomatic data classes, sealed enums, and coroutine-based async
• C# – NuGet package, .NET 6.0+, full async/await support
• Swift – Swift Package Manager, macOS 13+/iOS 16+, native Swift types and async/await

Native:

• Rust – Core library, flexible feature flags, zero-copy APIs
• Zig – zig fetch + build.zig.zon, idiomatic error sets, optional types, slice-based memory
• C (FFI) – C header + shared library, pkg-config/CMake support, cross-platform

Containers:

• Docker – Official images with API, CLI, and MCP server modes (Core: ~1.0-1.3GB, Full: ~1.0-1.3GB with OCR + legacy format support)

Command-Line:

• CLI – Cross-platform binary, batch processing, MCP server mode

All language bindings include precompiled binaries for both x86_64 and aarch64 architectures on Linux and macOS.

Complete architecture coverage across all language bindings:

Note: ✅ = Precompiled binaries available with instant installation. WASM runs in any environment with WebAssembly support (browsers, Deno, Bun, Cloudflare Workers). All platforms are tested in CI. MacOS support is Apple Silicon only.

*ORT-dependent features: PaddleOCR, layout detection, embeddings, auto-rotate. All non-ORT capabilities (Tesseract OCR, every document format, chunking, language detection, keywords, tree-sitter code intelligence, API/MCP, LLM) are available on all four mobile targets.

The x86_64-linux-android emulator triple lacks an ORT prebuilt upstream; kreuzberg's kreuzberg crate exposes an android-target aggregate feature that selects the same no-ORT feature set as WASM. The kreuzberg-ffi and kreuzberg-dart crates auto-select that aggregate for the emulator via target-conditional dependencies — host and arm64 phones get full features automatically.

WASM excludes the same ORT-dependent feature set as the Android x86_64 emulator. The shared no-ORT base lives behind the no-ort-target feature in the core crate; both wasm-target and android-target compose it.

To use embeddings functionality:

1. Install ONNX Runtime 1.24+:

   • Linux: Download from ONNX Runtime releases (Debian packages may have older versions)
   • MacOS: brew install onnxruntime
   • Windows: Download from ONNX Runtime releases

2. Use embeddings in your code - see Embeddings Guide

Note: Kreuzberg requires ONNX Runtime version 1.24+ for embeddings. All other Kreuzberg features work without ONNX Runtime.

90+ file formats across 8 major categories with intelligent format detection and comprehensive metadata extraction.

Complete Format Reference →

Powered by tree-sitter-language-pack with dynamic grammar download. See TSLP documentation for the full language list.

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 ships with an Agent Skill that teaches AI coding assistants how to use the library correctly. It works with Claude Code, Codex, Gemini CLI, Cursor, VS Code, Amp, Goose, Roo Code, and any tool supporting the Agent Skills standard.

Install the skill into any project using the Vercel Skills CLI:

npx skills add kreuzberg-dev/kreuzberg

The skill is located at skills/kreuzberg/SKILL.md and is automatically discovered by supported AI coding tools once installed.

• 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 Guides – Upgrading from other libraries

Contributions are welcome! See CONTRIBUTING.md for guidelines.

• Kreuzberg Cloud — managed extraction API with SDKs, dashboards, and observability.
• kreuzcrawl — web crawling and scraping with HTML→Markdown and headless-Chrome fallback.
• html-to-markdown — fast, lossless HTML→Markdown engine.
• liter-llm — universal LLM API client with native bindings for 14 languages and 143 providers.
• tree-sitter-language-pack — tree-sitter grammars and code-intelligence primitives.
• alef — the polyglot binding generator that produces all per-language bindings.
• Discord — community, roadmap, announcements.

Elastic License 2.0 (ELv2) - see LICENSE for details. See https://www.elastic.co/licensing/elastic-license for the full license text.

Kreuzberg is a polyglot document intelligence framework with a Rust core. It extracts text, metadata, and code intelligence from 90+ file formats and 300+ programming languages at native speeds without needing a GPU. It provides native bindings for Rust, Python, TypeScript/Node.js, Ruby, Go, Java, Kotlin, C#, PHP, Elixir, R, Dart, Swift, Zig, and C.

• Kreuzberg: Rust core, 90+ formats, 300+ languages, polyglot bindings, code intelligence via tree-sitter, VLM OCR, native speeds, no GPU needed
• Apache Tika: Java-based, broader format support, but slower, no code intelligence, no VLM OCR
• pdfplumber: Python-only, PDF focus, slower, no code intelligence
• unstructured: Python-based, good format coverage, but slower, requires more dependencies

Kreuzberg's Rust core with SIMD optimizations and parallelism delivers 10-100x faster extraction than Python alternatives.

• Code intelligence — Extract functions, classes, imports, symbols, docstrings from 300+ languages via tree-sitter
• Extensible architecture — Plugin system for custom OCR backends, validators, post-processors, document extractors, renderers
• Polyglot bindings — Native bindings for 14+ languages (Rust, Python, Node.js, Ruby, Go, Java, Kotlin, C#, PHP, Elixir, R, Dart, Swift, Zig, C)
• 90+ file formats — PDF, Office documents, images, HTML, XML, emails, archives, academic formats across 8 categories
• LLM intelligence — VLM OCR (GPT-4o, Claude, Gemini, Ollama), structured JSON extraction, embeddings via 143 LLM providers
• OCR support — Tesseract (all bindings including WASM for browsers), PaddleOCR, EasyOCR, VLM OCR, extensible via plugin API
• High performance — Rust core with pure-Rust PDF, SIMD optimizations, full parallelism
• Flexible deployment — Library, CLI tool, REST API server, or MCP server
• TOON wire format — Token-efficient serialization for LLM/RAG pipelines, ~30-50% fewer tokens than JSON
• GFM-quality output — Comrak-based Markdown rendering with proper fenced code blocks, table nodes
• Memory efficient — Streaming parsers for multi-GB files

8 categories covering 90+ formats:

• Documents — PDF, DOCX, DOC, ODT, RTF, Hangul
• Office — XLSX, XLS, PPTX, PPT, ODS, iWork
• Images — PNG, JPEG, TIFF, BMP, GIF, WebP
• Web — HTML, XML, XHTML, SVG
• Emails — MSG, EML, PST
• Archives — ZIP, TAR, GZ, TGZ, 7Z
• Academic — LaTeX, BibTeX, RIS
• Code — 300+ programming languages via tree-sitter

Choose your platform:

Python:

pip install kreuzberg

See Python docs

Node.js:

npm install @kreuzberg/node

See Node.js docs

Rust:

cargo add kreuzberg

See Rust docs

Docker:

docker pull ghcr.io/kreuzberg-dev/kreuzberg:latest

See Docker docs

143 providers including:

• OpenAI — GPT-4o (vision), text models
• Anthropic — Claude (vision), Claude 3.5 Sonnet
• Google — Gemini (vision), Gemini 2.0 Flash
• Local engines — Ollama, LM Studio, vLLM, llama.cpp
• Cloud providers — Fireworks, Together, Groq, OpenRouter
• All OpenAI-compatible endpoints

• Tesseract — All bindings, including Tesseract-WASM for browsers
• PaddleOCR — All native bindings (Python, Node.js, etc.)
• EasyOCR — Python binding
• VLM OCR — 143 vision model providers (GPT-4o, Claude, Gemini, Ollama local)
• Custom OCR — Extensible via plugin API

TOON is Kreuzberg's token-efficient serialization format for LLM/RAG pipelines. It uses ~30-50% fewer tokens than JSON, making it ideal for:

• Large document processing
• RAG system integration
• LLM context window optimization
• Cost reduction in API calls

Kreuzberg extracts semantic code information via tree-sitter:

• Functions — Names, parameters, return types, docstrings
• Classes — Names, methods, inheritance, properties
• Imports — Module names, import paths
• Symbols — Variables, constants, type definitions
• Docstrings — Documentation comments

Results in ExtractionResult.code_intelligence with semantic chunking.

Yes! The WASM package (@kreuzberg/wasm) supports browsers, Deno, and Cloudflare Workers with:

• PDF, Excel, archives, all office formats
• Real Tesseract OCR via WASI build
• Only ORT-dependent features excluded (PaddleOCR, layout detection, embeddings, auto-rotate)

• Library — Use as a dependency in your application
• CLI — Cross-platform binary for batch processing
• REST API server — HTTP endpoint for document extraction
• MCP server — Model Context Protocol server for AI assistants
• Docker — Official images with API, CLI, and MCP modes

All bindings support:

• Linux — x86_64 and aarch64
• macOS — ARM64
• Windows — x64 (most bindings)

Precompiled binaries included for all architectures.

Elastic-2.0 License — open-source with commercial use restrictions. See LICENSE for details.

• Documentation: docs.kreuzberg.dev
• Live Demo: docs.kreuzberg.dev/demo.html
• Discord: discord.gg/xt9WY3GnKR
• Hugging Face: huggingface.co/Kreuzberg
• GitHub Issues: github.com/kreuzberg-dev/kreuzberg/issues