dsh-plugin-multimodal

In DeepSeek Harness, the official setup only supports plain text input, and pasting an image results in a 'model does not support images' error. This creates inconvenience for users who need to reference screenshots, charts, or other visual content. This plugin addresses this pain point by providing image admission functionality, ensuring that pasting images is never rejected by the system.

Its core capability lies in intelligently adapting to different main models: if the current model natively supports vision (like Claude or GPT), the plugin passes the image directly to the model for processing; if the main model is text-only (such as the official DeepSeek), the plugin uses a configurable visual sidecar to convert the image content into a text description, which is then passed to the main model. This approach maintains compatibility while seamlessly enabling image support.

This plugin is ideal for all users who need to handle image content within DeepSeek Harness, especially those relying on text-only models but occasionally needing to analyze error screenshots, web designs, or data charts. Whether you are a developer, researcher, or everyday user, as long as you want to include image elements in conversations without disruption, this plugin offers a lightweight solution. It focuses solely on making pasting possible rather than building a complex visual toolkit, making it perfect for scenarios that prioritize simplicity and efficiency.