dsh-eyes

When using DeepSeek Harness for text-only conversations, have you ever faced the awkwardness of needing to "see" an image but only being able to "read" text? dsh-eyes is designed to solve this pain point. It acts like adding a pair of "eyes" to your chat interface, capable of intelligently intercepting images pasted, dragged, or imported, converting them into text content, and seamlessly integrating them into the dialogue flow. Whether it is a complex screenshot or a document, it can quickly transform it into readable text, bridging the gap between images and text models.

The plugin's core capability lies in its flexible and powerful interaction modes. You can choose to enable image interception, allowing the plugin to automatically recognize thumbnails and extract text (supporting up to 9 images at once), or disable interception to send images directly to multimodal models for real-time analysis. It also supports the vision_ask tool, which combines recognition results with your instructions, allowing the main model to ask questions and refine details based on understanding. Additionally, the plugin includes various model presets and supports directly reading local file paths to adapt to various workflow needs.

Whether you are a researcher needing complex document analysis or a visual worker accustomed to mixed text and graphics, dsh-eyes is a tool to boost efficiency. It not only simplifies the interaction between images and text but also provides powerful toolchain support, enabling large language models to not only "see" but also engage in deep follow-up questions and detail correction based on understanding.