dsh-pdf-reader

Reading technical papers, textbooks, or multiple PDFs alongside a DeepSeek Harness conversation usually means bouncing between a browser reader, an annotation tool, and the AI chat window. dsh-pdf-reader collapses that loop into a single web workflow: you can paginate, select text, drag in image regions, and ask questions without ever leaving your current context.

The plugin follows a read-select-ask-annotate-save pipeline. It ships with multi-PDF tabs, table of contents, bookmarks, full-text search, and lossless rotation. Once you select a passage you can copy, explain, translate, or drill deeper, and highlight, underline, strikethrough, and note annotations are managed in a unified annotation panel. Writing to the PDF only happens when you choose the save-as option or confirm overwrite. The standout feature is multimodal joint analysis: add figures, captions, and related paragraphs into a shared material basket, and the AI reasons across documents and media, returning source links that jump back to the original page. OCR is off by default; after you explicitly authorize it, the AI picks only the pages it needs within a configurable budget, so an entire book is never scanned automatically.

If your daily work involves reading technical documentation with AI assistance, cross-referencing multiple PDFs, or annotating while you interrogate the text, this plugin removes most of the context-switching overhead. The current build targets DeepSeek Harness 0.1.1-rc.2, so it is wise to verify the save flow on non-critical PDF copies first.