dsh-vision-plugin

DeepSeek Harness typically powers pure-text models, which limits scenarios requiring image processing, such as visual Q&A, image analysis, or content creation. The dsh-vision-plugin elegantly addresses this by leveraging free vision models from OpenRouter to add image understanding capabilities to pure-text models like DeepSeek—at no cost. Its core features include automatic handling of pasted or uploaded images, converting visual content into text descriptions for the model, and offering a describe_image tool to directly read local image files. The plugin also intelligently adapts to model capabilities: if the current model supports images, they are sent as-is; otherwise, conversion happens behind the scenes to keep conversations smooth. It's ideal for developers, researchers, or everyday users who want to expand DeepSeek Harness's multimodal applications with zero overhead, such as quickly analyzing charts, recognizing image content, or building smarter chatbots.