v1.0.8

Highlights

• Candle OCR backends in published builds. The Candle VLM backends (TrOCR, PaddleOCR‑VL, GLM‑OCR, DeepSeek‑OCR) are compiled, registered, and clippy‑checked in the published crates/wheels/packages, so candle-*-ocr backends resolve out of the box (except wasm/android/dart/swift‑mobile). Model weights download from HuggingFace on first use.
• Durable Swift artifact bundle. The swift‑bridge shim's injected xberg-ffi dependency is now generated with default-features = false + full-no-heic (via alef 0.49.0's new ffi_features knob), so heic's libheif-sys no longer breaks cross‑compilation — no post‑regen hand‑patch. Also fixes an iOS dead‑code build error in xberg-libwpd under -D warnings.
• #1355 — force_ocr no longer silently drops pages. When pdf-oxide cannot rasterize an image XObject, extraction now OCRs the raw embedded image and records a ProcessingWarning instead of emitting a blank page.
• Windows XLSX extraction is enabled in the Windows target.
• npm integrations fixed & version‑synced. langchain/llamaindex loaders use the ExtractInputKind enum; the n8n node handles the readonly config types; all integration manifests pin 1.0.8.

Extraction quality

Numerous structural/quality fixes: Typst headings & tables, Org Mode tables & headerless rows, EPUB switch semantics, DOCX structure, image/TATR layout & performance, MSG email body decoding, XLSX performance, and table edge cases.

Notes

• crates.io, PyPI, npm (+ integrations), and Docker are published from this tag.
• Internally, bindings are regenerated with alef 0.49.0, which adds the SwiftConfig.ffi_features codegen knob that makes the Swift build fix regen‑stable.

Zig

Add to your build.zig.zon:

.dependencies = .{
    .xberg-zig = .{\n        .url = \"https://github.com/xberg-io/xberg/releases/download/v1.0.8/xberg-zig-v1.0.8.tar.gz\",\n        .hash = \"xberg-1.0.8-iV1Grn-xBRfohNsypN9wjJfvyjT2rYCikrrGio2mXpqR\",\n    },\n},\n```\n