Xberg is an extremely fast document intelligence engine, written in Rust with bindings for more languages (Python, Node, Go, Java, C#…). It extracts and structures text, tables and metadata from over 100 file formats — PDFs, Office and OpenDocument files, spreadsheets, presentations, images, e-mail and more — fully local and offline, with OCR fallback (Tesseract, Paddle) and content-hash caching.
Xberg is the successor of the open source library Kreuzberg and takes over its installation base with a new v1 release line. It can be used as library, as REST service, as CLI or as MCP server, which makes it easy to integrate it into GenAI systems: reliable local document parsing is the prerequisite for feeding documents into embeddings and retrieval pipelines.
We run Xberg as a container in MUCGPT 2.0 to enable document upload and parsing in chats ("Chat with your Docs") and use it in Zammad-AI to extract ticket attachments for retrieval.