wtf is markitdown-vellum?

markitdown-vellum is a plugin for Vellum Assistant that lets an AI model read office documents, PDFs, spreadsheets, presentations, notebooks, and archive files as plain markdown text, without a person having to open or copy the content by hand. It is built on top of an existing library called cognipeer to-markdown. The plugin adds a single tool called markitdown to the assistant. When the model points this tool at a file in the workspace, it gets back extracted text, including headings, tables, slide content, spreadsheet data, email bodies, and descriptions of images where that applies. Scanned PDFs and images are handled through optical character recognition, either by using the workspace's own vision model or by using a bundled version of Tesseract, a well known open source OCR engine. The README is clear about scope: plain text files and audio files are not supported, since the underlying library's list of supported file types leaves them out. Supported file types otherwise follow whatever the cognipeer to-markdown library supports. Installation can be done through the Vellum plugin catalog or by pulling the code from GitHub. For local development, you copy the plugin folder into the workspace's plugins directory, run npm install, and reload the plugins. Configuration happens through a config.json file in the plugin folder that survives upgrades, with settings for how long a single file conversion is allowed to take, which OCR mode to use, and what prompts get sent to the vision model during OCR. No extra system dependency is needed for the Tesseract mode, since that library is bundled in, and PDF page rendering uses a canvas module that installs automatically.