Xberg v1, Rust-based AI extraction released

Xberg v1 has been released as an open-source content intelligence framework that extracts and refines content from documents, code and data, audio and video, and static and dynamic URLs. It is the successor to the existing Kreuzberg v4.

Key changes are as follows.

• Introduced a pure Rust PDF backend pdf_oxide, removing the native pdfium dependency.
• Restores document reading order by combining ONNX-based layout detection with predecessor graph reordering.
• Determines whether each page is scanned to run OCR only on the pages that need it, and also supports AcroForm/XFA fields and titles based on the PDF table of contents.
• Supports PaddleOCR PP-OCRv6 and Tesseract, and provides a pure Rust Candle inference stack for TrOCR, GLM-OCR, GOT-OCR, DeepSeek-OCR, and PaddleOCR-VL.
• Supports WASM browser and mobile inference via a tract-based ONNX Runtime-independent inference path.
• Provides Rust-native GLiNER2 named entity recognition and a WASM model for browsers.
• Handles 101 document formats and 367 code and data formats, and also optimizes memory usage, model sessions, and streaming conversion for OCR and PDF extraction.