Xberg 1.0 released: document extraction for a world of tooling
16 hours ago
- Xberg 1.0 is a document extraction engine written in Rust, offering 15 generated language bindings and a flexible data model.
- It is a rebrand of Kreuzberg; migration requires changing project coordinates, but legacy support continues until 2026.
- Xberg's output model includes elements, tables, images, chunking, and structured extraction, designed for AI pipelines rather than just text.
- In benchmarks against Apache Tika, Xberg was roughly 3–5 times faster but had a slower initialization time (1.5s vs 0.4s).
- Correctness was comparable, with Xberg failing on one pathological DOCX and one GIF with inverted text (polarity bug, later fixed).
- WordPerfect support was added rapidly by maintainers, demonstrating high development velocity.
- Xberg reduces dependency count from 86 JARs to 5 but increases payload size to 199 MB due to bundled native libraries.
- Overall, Xberg is a fast, modern extraction tool suited for agent-oriented document processing.