dsh-document-reader

When chatting with an AI in the DeepSeek Harness web UI, you often need to feed a PDF report, a Word document, or a PPT deck into the prompt. Opening the file, selecting all the text, copying, and pasting into the composer breaks your flow and can silently drop footnotes or table cells. dsh-document-reader turns that into a two-step action: click the Document button next to the input field to pick a file, or simply drag the file anywhere on the page and the extracted text lands in the draft instantly.

The plugin handles .pdf (parsed server-side with pdfjs-dist, correctly processing object streams and CID fonts), .docx, and .pptx (unpacked client-side via ZIP + XML), as well as plain-text formats such as txt, md, csv, json, yaml, and source code. Each file is capped at 200 k characters; anything beyond is truncated with a note. Image-only scanned PDFs and legacy binary .doc / .xls / .ppt files are not supported save them to their modern equivalents first.

If your workflow regularly involves handing documents to an AI for summarization, translation, code review, or Q and A, this plugin removes the copy-paste friction and keeps the conversation moving. A modern browser (Chrome, Edge, or Firefox 2022 or later) is required for the client-side parsing.