dsh-upload-plugin

The DeepSeek Harness chat bar originally lacked a natural attachment entry point. Asking a vision model to examine an image or a text model to read a PDF usually meant leaving the conversation and fumbling with file paths outside the chat. dsh-upload-plugin closes that gap in the plus-menu: Add photos lets you select multiple images at once, Add files accepts PDFs, code files, archives and more, and the Uploaded files view lets you browse everything sent in the current session and re-analyze it with a single click.

Attachment cards sit natively inside the chat capsule. Images render as 64x64 rounded thumbnails with a full-screen lightbox on click; file cards display a format badge, name, and size with a subtle hover-to-reveal dismiss button. All uploads are scoped to their session so switching conversations never leaks files across sessions. Images over 4.5 MB are automatically compressed to stay under the read_image tool limit, and the plugin checks whether the active model declares image input capability before upload, giving text-only models a clear warning instead of a silent failure.

Built for developers in DeepSeek Harness who regularly feed images or documents into their models, especially those using Custom Providers with image modality configured. Once installed, no manual directory management or config edits are needed: the plus-menu makes sharing a picture or a file as effortless as typing a message.