// the find
xberg-io/xberg
A polyglot document intelligence framework with a Rust core. Extract text, metadata, images, and structured information from PDFs, Office documents, images, and 97+ formats. Available for Rust, Python, Ruby, Java, Go, PHP, Elixir, C#, R, C, TypeScript (Node/Bun/Wasm/Deno)- or use via CLI, REST API, or MCP server.
Xberg is a Rust-core document extraction library that pulls text, tables, metadata, and structured data from 96 file formats, with native bindings for 15 languages and multiple deployment modes. It's a rebranded continuation of Kreuzberg, aimed at developers building RAG pipelines, document processing services, or anything that needs to reliably crack open PDFs, Office files, and images at scale. The Rust core means it's fast and memory-safe; the binding breadth means your language of choice is probably covered.
The polyglot binding story is genuinely impressive — Go, Rust, Python, Node, C#, Java, Swift, Dart, Zig, and more, all apparently generated from a single binding tool (alef). The caching layer using content-hash keys is the right call for production workloads where the same document gets submitted repeatedly. The OCR backend abstraction with pluggable fallback chains (Tesseract → PaddleOCR → Candle → VLM) means you can tune cost vs. accuracy without rewriting your integration. MCP server support is a practical addition that makes it drop-in compatible with Claude Desktop and similar AI coding tools.
This is v1 of a rebranded project, and the README's scope is so wide it reads like a features-not-shipped list — 96 formats, 306 languages, 15 bindings, embeddings, NER, transcription, token reduction, structured LLM extraction. That's not a library, it's a platform, and first versions of platforms usually have several weak links. The audio/video support is transcription-only (audio track extracted, fed to Whisper), which is fine but the table implies broader video intelligence. The `.ai-rulez/` directory with agent rules baked into the repo is a non-obvious governance pattern that will confuse contributors who aren't using AI coding tools. No changelog for the v1 line is visible in the tree, making it hard to know what actually changed from Kreuzberg.