Self-hostable
document intelligence.
The extraction, OCR, chunking and embedding primitives behind Xberg — published as xberg, MIT-licensed and free to run anywhere.
Everything the pipeline is made of. In the open.
Extraction
Text, tables and metadata out of 96 file formats — PDFs, Office documents, images, email and more.
OCR
Tesseract-backed OCR for scans and photos, with vision-language models available as a backend.
Chunking
Semantic chunking across prose, markdown and code, with keyword extraction and rich metadata.
Embeddings
Ultra-fast embeddings via a Rust-native ONNX engine. Four presets, extensible to any model.
Code intelligence
Functions, classes, imports and symbols extracted across 306 programming languages.
Polyglot SDKs
Fourteen language bindings — Python, TypeScript, Rust, Go, PHP, Ruby, C#, WASM and more.
Six focused libraries. Each useful on its own.
Xberg is composed of small, single-purpose open-source projects. Take the whole pipeline, or pull in just the piece you need.
The extraction core — text, tables and metadata from 96 formats, with optional OCR.
Web crawling & scraping, with HTML→Markdown and a headless-Chrome fallback.
A fast, lossless HTML→Markdown engine.
One LLM client across 14 languages and 143 providers.
Tree-sitter grammars and code-intelligence primitives across 306 languages.
The polyglot generator that produces every per-language binding.
Built with Xberg.
LegalRAG
Contract-review assistant that ingests thousands of scanned PDFs and answers clause-level questions.
PaperMind
Chat over arXiv papers — tables and figures preserved, chunked and embedded end to end.
InboxIQ
Turns messy email attachments into structured records for an operations team.
ShelfScan
Invoice and receipt extraction pipeline powering a small accounting SaaS.
CodeAtlas
Semantic code search across a monorepo using the 306-language tree-sitter parser.
DocuChat
Self-hosted chat-with-your-docs, with OCR for scanned files built in.
Built something on the open-source primitives? Share it on Discord and we'll feature it here.
MIT licensed. No fine print.
Free for anything
Personal, internal and commercial use are all free. Run it in production, ship it inside your product, sell what you build.
Self-host anywhere
Your infrastructure, your data. Nothing phones home — documents never leave your machines.
Just keep the notice
MIT's only ask is that the copyright notice travels with the code. Fork it, embed it, build a business on it.
Fits into
your workflow.
Drop-in integrations for the AI frameworks and tools you already build with.
from langchain_xberg import XbergLoader# Load any document as LangChain Documentsloader = XbergLoader("report.pdf")docs = loader.load()# Drop straight into your vector storedb.add_documents(docs)
One API.
Every language.
Install Xberg in your stack and call the same extraction API — from Python to Rust to the browser.
12 languages
Python, TypeScript, Rust, Go, Java, Ruby, and more.
306 code formats
Functions, classes, imports, symbols — all parsed.
Docker & CLI
Run our images or the single-binary CLI.
ELv2 licensed
Free for personal, internal, and commercial use.
Questions developers ask.
- How is the open-source library different from Xberg Enterprise?
- The library is the extraction engine you run yourself — OCR, chunking, embeddings, code intelligence. Xberg Enterprise wraps that same engine in a managed, multi-tenant platform: REST API, dashboards, metered billing, signed webhooks, observability, and Kubernetes deployment. Same core underneath; the difference is the operational plumbing around it.
- Is it really free for commercial use?
- Yes. It's MIT licensed — free for personal, internal and commercial use. Ship it inside your product, run it in production, build a business on it. The only requirement is that the copyright notice travels with the code.
- Do I need an account or API key?
- No. The open-source library runs entirely on your own machines — no account, no key, no sign-up. An API key is only for the managed cloud service.
- Does my data leave my machine?
- No. Self-hosted means documents never leave your infrastructure and nothing phones home. You choose your own OCR, embedding and LLM backends, including fully local models with no keys to configure.
- Which languages can I call it from?
- Fourteen. Native bindings for Python, TypeScript/JavaScript, Rust, Go, PHP, Ruby, C# and more — all generated from one core, so behaviour stays consistent across languages.
- Does OCR need extra setup?
- Basic OCR uses Tesseract; install the language packs you need. For higher accuracy you can plug in vision-language models — including self-hosted GPU models — as the OCR backend.
- Is it production-ready and actively maintained?
- Yes — it's the same engine that powers the Xberg cloud in production. Development happens in the open on GitHub, with regular releases across all fourteen language bindings.
- How do I get help or contribute?
- Open an issue or pull request on GitHub, or join the Discord to ask questions and share what you're building.