Xberg

Xberg pulls clean text, tables, metadata, and structured data out of nearly any file you point it at: PDFs, scanned images, Office documents, audio, archives, even a whole source tree. One Rust engine does format detection, reading, OCR, and extraction, so you stop stitching six libraries into a pipeline. MIT licensed, and the project states it plainly as free for personal, internal, and commercial use.

Two things changed in June 2026 that matter if you looked at this before. The project was called Kreuzberg and it spent about eleven weeks under the Elastic License, which blocked offering it as a service. On 2026-06-25 it rebranded to Xberg and relicensed back to MIT on the same day. Deployment is a Docker image, a CLI, a REST server, a Helm chart, or an MCP server, with bindings for fifteen languages.

Solo builders wiring up a RAG pipeline get the obvious win: one dependency instead of a pile. Teams doing search indexing or document ingestion get 100 plus formats and 371 code languages behind a single API. Docling (docling-project/docling) is the closer comparison when your inputs are mostly PDFs and you want IBM's layout models.

The catch: the commercial side is deliberately closed. The only paid offering is an invitation-only Design Partner program with a managed backend and pricing you find out during the application call. That does not restrict the MIT code at all, but it does mean there is no ordinary support contract to buy when self-hosting goes sideways.