Kreuzberg is a high-performance, polyglot library designed to extract text and metadata from over 57 file formats, including comprehensive OCR capabilities. Built with a Rust core, it offers native speed processing, memory efficiency, and the ability to generate embeddings without requiring a GPU, making it highly versatile for various data extraction and processing tasks.

kreuzberg

Active·★ 9.5k·MIT·Updated 2026-10-27

★ Essential★ Trending

Kreuzberg extracts text and metadata from a wide range of file formats (57+), generates embeddings and post-processes the results at native speed without needing a GPU. The core is written in Rust with native PDFium, SIMD optimisations and full parallelism, and streaming parsers keep memory use down for multi-gigabyte files.

Native bindings exist for Rust, Python, TypeScript and Node.js (including a WebAssembly build for browsers, Deno and Cloudflare Workers), Ruby, Go, Java, C#, PHP and Elixir, and the same engine can be used as a library, a CLI, a REST API server or an MCP server. Formats span PDF, Office documents, images, HTML, XML, email, archives and academic formats. OCR is available through Tesseract (all languages via the native binding), EasyOCR and PaddleOCR in Python, and Guten in Node.js, and a plugin system allows custom OCR backends, validators, post-processors and extractors. Optional embeddings support is included, and the project is MIT licensed.

#Document Processing#Data Extraction#OCR#Multi-language#Embeddings#Coding#Data Analysis#Image Generation

Features

01Extensible architecture with a plugin system for custom backends and processors.

02Polyglot support with native bindings for 10+ programming languages.

03Comprehensive support for 57+ file formats across 8 categories, including Office, PDF, and images.

04Advanced OCR capabilities with multiple backends and intelligent table detection.

05High performance due to a Rust core, SIMD optimizations, and full parallelism.

Why choose it

+One engine reads 57+ formats (PDF, Office, images, HTML, email, archives, academic) and runs without a GPU.

+Native bindings for nine languages plus CLI, REST and MCP server, so it fits most stacks.

+Rust core with streaming parsers keeps speed and memory use reasonable on multi-gigabyte files.

+Plugin system for custom OCR backends, validators, post-processors and extractors.

Trade-offs

!Platform support is not uniform: the README's table shows no Windows build for the Ruby binding.

!OCR options differ by language (EasyOCR and PaddleOCR are Python-only, Guten is Node.js-only), so results depend on the binding you choose.

!Embeddings are optional and need extra setup, and the extraction quality depends on the document type and OCR backend.

Maintenance check

ActiveChecked: 2026-10-25

Last commit: 2026-10-25
Commits, last 90 days: 100+
Releases, last 90 days: 30
Latest release: v1.3.10 (2026-10-13)
Open issues: 22
Recent issues with at least one comment: 12/23

Commits per week, last 12 weeks

Oldest week on the left; the latest four weeks are orange. Source: GitHub commit activity.

Status rule: archived on GitHub = Archived; last commit within 90 days = Active; 91 to 365 days = Slowing; older = Stale. All figures come from the GitHub API on the date shown.

Most discussed open issues

#1567 chore(lint): pay down the size and complexity baseline14 comments · 2026-09-05
#2132 bug(ocr): VLM transcription drops repeated body paragraphs6 comments · 2026-10-16
#2126 release: prepare 1.4.0 after security, runtime and artifact gates4 comments · 2026-10-16
#2236 ci: workflows build an unreleased ai-rulez that does not match the named 5.3.01 comments · 2026-10-24
#2182 HTML extraction panics on whitespace-only inline formatting (xberg 1.3.5–1.3.10, html-to-markdown-rs 3.17)1 comments · 2026-10-20

Issue titles are quoted from GitHub. Open the link to check whether a problem still applies to the version you use.

Who it fits

Pick it if

You are building an ingestion pipeline and need text and metadata from 57+ formats without a GPU.
Your stack is Rust, Python, TypeScript, Ruby, Go, Java, C#, PHP or Elixir, or you need a WebAssembly build.
You process multi-gigabyte files and need streaming parsers, or want custom OCR backends through plugins.

Look elsewhere if

You just want an assistant to read a file in chat: a simple MCP converter is quicker to set up.
You need the Ruby binding on Windows: the README's table shows no Windows build for it.
You need the same OCR everywhere: EasyOCR and PaddleOCR are Python-only and Guten is Node.js-only.

How it differs

vs anymd★ 1.1k

anymd is built for agents reading documents through MCP. Kreuzberg is a library and server for extraction at scale, with MCP as one of several interfaces.

Full comparison →

Compatibility

Rust

Core Library

Verified via docs

Python

Language Binding

Verified via docs

Elixir

Language Binding

Verified via docs

Node.js

Language Binding

Verified via docs

WASM

WebAssembly Support

Verified via docs

Java

Language Binding

Verified via docs

Quick start

$ pip install kreuzberg

Use cases

↳Automated extraction of text, metadata, and structured data from diverse document types.

↳Building intelligent document processing pipelines for data ingestion and analysis.

↳Enabling efficient search and retrieval systems for unstructured and semi-structured content.

Alternatives

⚡Top VS:

VSanymd★1.1k→VSMeshroom★13.0k→VSCodeFuse-muAgent★777→VSai-doc-gen★767→

anymd★ 1.1k

Any file → clean Markdown for AI agents: PDF, Word, PowerPoint, Excel, EPUB, HTML and web pages, images (OCR), audio and video (metadata, subtitles, transcripts). A fast Rust MCP server and CLI that runs on your machine. No API key.

Features

Why choose it

Trade-offs

Maintenance check

Most discussed open issues

Who it fits

How it differs

Compatibility

Quick start

Use cases

Alternatives

Featured in these lists

Comments

Features

Why choose it

Trade-offs

Maintenance check

Most discussed open issues

Who it fits

How it differs

Compatibility

Quick start

Use cases

Alternatives

Featured in these lists

Comments