Kreuzberg (Xberg): Document Intelligence & PDF Extraction

Overview

Kreuzberg (now part of the Xberg ecosystem) is a polyglot document intelligence framework built with a Rust core. It extracts text, metadata, images, and structured information from over 97 file formats — PDFs, Office documents, images, EPUB, HTML, and more. If you’ve ever fought with a PDF library that couldn’t handle a simple table or mangled your Chinese characters, Kreuzberg is the relief you’ve been looking for.

The project started as a standalone skill for AI coding agents and grew into a full framework with bindings for Rust, Python, Ruby, Java, Go, PHP, Elixir, C#, R, C, and TypeScript (Node/Bun/Wasm/Deno). You can also use it via CLI, REST API, or MCP server. The Rust core means it’s fast — we’re talking milliseconds per page on modern hardware.

Why It Matters for AI Workflows

Every RAG pipeline, every document processing workflow, every “upload this PDF and ask questions about it” feature — they all hit the same wall: extracting clean text from messy documents. Kreuzberg solves this at the infrastructure level.

Key capabilities:

• PDF extraction — text, images, embedded tables, annotations, and metadata
• OCR pipeline — Tesseract integration for scanned documents and image-heavy PDFs
• Table extraction — preserves row/column structure instead of dumping a wall of text
• 97+ format support — DOCX, XLSX, PPTX, ODT, EPUB, HTML, Markdown, and more
• WASM build — runs directly in the browser with no server needed

Performance You Can Count On

The Rust core handles heavy lifting while the bindings give you a familiar API in whatever language you’re working in. In benchmarks against popular Python libraries (PyMuPDF, pdfplumber, pdfminer), Kreuzberg consistently delivered 2-5x faster extraction with better accuracy on complex layouts.

Scanned documents? The built-in OCR pipeline detects image-based pages automatically and routes them through Tesseract. No manual switching between text and OCR modes.

Getting Started

# Python (most common for AI workflows)
pip install kreuzberg

# CLI usage
kreuzberg extract document.pdf --output text

# Python API
from kreuzberg import extract_text
result = extract_text("annual-report.pdf")
print(result.text)

For AI agent integration, the MCP server is the cleanest path:

npx @xberg/mcp-server
# Now your AI agent can read any document format natively

Who Should Use It

• RAG pipeline builders — replace brittle PDF parsers with something that actually works
• AI agent developers — give your agents the ability to read real-world documents
• Data extraction teams — batch process thousands of documents with consistent output
• Full-stack devs — the WASM build means browser-only document processing

Final Thoughts

Kreuzberg is one of those tools that quietly solves a massive pain point. Document extraction isn’t glamorous, but every AI developer hits it sooner or later. Having a single framework that handles 97+ formats with a Rust-speed core and bindings in every major language is a massive time saver. The project has 8,600+ GitHub stars and growing fast — the community is active, and the maintainers ship updates regularly.