dsh-docs

For DSH agents that need to work with documents, the first hurdle is often not summarizing content but turning PDFs, Office files, and scanned images into reliable formats the model can consume. dsh-docs handles that locally: it lets DeepSeek Harness read files inside explicitly authorized directories and return clean Markdown, plain text, or structured JSON.

Its core strength is offline document parsing plus local OCR. On Windows x64 it can use a self-contained Python and Xberg runtime with local Tesseract data for English and Simplified Chinese, while the Node/Xberg path remains a lightweight non-OCR fallback. Supported inputs include PDF, DOCX, XLSX, PPTX, Markdown, HTML, CSV, text, PNG, JPEG, TIFF, WebP, and scanned PDFs.

It is especially useful for privacy-conscious users and workflow builders. Documents stay on the local disk, and the plugin needs no Docker, HTTP service, or API keys; remote URLs are detected only to reject them. Researchers, developers, knowledge workers, and DSH agent builders can use dsh-docs as a stable local document preprocessing layer.