dsh-guide-dog

DeepSeek Harness is a powerful inference framework, but it lacks native visual input, limiting its ability to process images directly. The Guide Dog plugin bridges this gap by integrating MiniMax's multimodal services, giving DeepSeek 'eyes' and 'hands' to describe images, review designs, generate media, and interact via voice. Its core capabilities include VLM-based image description and structured inspection, multi-type content generation (images, videos, speech, music, text), inline Web UI preview and playback, and real-time voice conversation modes. Through automatic invocation of visual check tools, it ensures models handle visual tasks correctly even when the base model cannot see images.

This plugin is ideal for developers using DeepSeek Harness for content creation, frontend development, or multimodal tasks. Whether you need to review screenshots, generate promotional materials, or interact with an assistant via voice, Guide Dog offers seamless integration. It also lets users customize voice mode, input devices, and other preferences through a settings page, while planned accessibility features will further expand its reach. In short, Guide Dog transforms DeepSeek Harness from a text-only inference tool into a versatile assistant that can see, hear, speak, and create.