Xberg, 101-format extraction engine

Xberg is a Rust-based document intelligence engine that extracts text and tables from 101 formats, including PDFs, office documents, images, HTML, emails, e-books, academic papers, and structured data.

It handles format detection, document reading, OCR, and structured extraction in a single core, and can be used in the following ways:

• 15 language bindings including Python, Node.js, Go, Java, C#, Ruby, PHP, and Elixir
• Library, CLI, REST API server, MCP server
• Structured output including plain text, Markdown, Djot, HTML, JSON tree, and OCR metadata/bounding boxes

It also supports the HWP·HWPX formats, which are widely used in domestic document processing. When a code repository is provided, it extracts functions, classes, imports, symbols, and docstrings from 371 programming languages based on tree-sitter, and allows chunking aligned with syntactic boundaries.

According to the full-corpus p50 comparison in the benchmark released by Xberg, it recorded a throughput of 1.70MB/s for Markdown output, a cold start of 203ms, a success rate of 100%, and a maximum memory usage of 32MB. However, Docling achieved a higher layout score of 0.884, and these performance figures are from Xberg's own benchmark results.