kreuzberg-dev/kreuzberg-lts

Kreuzberg is a tool that pulls text, metadata, and code structure out of a very wide range of file formats, including PDFs, Office documents, images, HTML, emails, and archives, plus source code in hundreds of programming languages. It is built with a Rust core for speed and does not need a graphics card to run. This particular repository is the long term support version of Kreuzberg 4, meaning it still gets bug and security fixes through the end of 2026, but new features are now being added to a separate successor project instead. For code, Kreuzberg can identify functions, classes, imports, and other symbols across roughly 248 programming languages using a parsing technique called tree sitter, and it can break documents into meaningful chunks. For documents, it covers more than 90 file formats grouped into eight categories, such as PDFs, spreadsheets, images, and academic paper formats. It can also run optical character recognition using several different engines to pull text out of scanned images, and this can be extended further through a plugin system. The project offers native bindings for many programming languages, including Rust, Python, TypeScript and Node.js, Ruby, Go, Java, C sharp, PHP, Elixir, R, and C, so a developer can use whichever language their own project is already written in. It can also connect to external AI language models for tasks like reading scanned documents with a vision model or pulling out structured JSON data, and it supports over a hundred different AI providers for that purpose. Kreuzberg can be run as a code library inside another program, as a command line tool, as a web server offering a REST API, or as an MCP server for AI coding agents. It is also available as a Docker container. The output can be produced in several formats such as Markdown, HTML, or plain text. The project is released under the MIT license, which allows free use including in commercial projects.