dsh-vision-provider

DeepSeek Harness enables users to select a single model for sessions, but initially made it challenging to use vision models alongside DeepSeek's reasoning capabilities, limiting scenarios that require image processing and text generation. The dsh-vision-provider plugin solves this by integrating selectable vision model combinations directly into the Web UI, allowing users to seamlessly incorporate vision models as part of a DeepSeek provider. Image-bearing messages are analyzed by the vision model, while DeepSeek handles reasoning, tool use, and final response generation based on the visual analysis, creating a smooth two-model bridge. This plugin is ideal for developers building applications that merge visual analysis with natural language processing, or any user seeking to process images and generate intelligent responses within DeepSeek Harness. It is particularly suited for tasks like document analysis, image captioning, and other vision-related workflows, where efficient integration of visual and textual capabilities is essential.