GitHub radar
DSH Vision Toolkit: 10 AI Vision Tools for DeepSeek Agents
DSH Vision Toolkit is an open source DeepSeek Harness plugin that brings 10 visual capabilities — image Q&A, OCR, UI reconstruction from a screenshot, etc.—via free Gemini 3.7 Flash without any API key.
DSH Vision Toolkit is a plugin of DeepSeek Harness that brings 10 visual features such as image Q&A, pixel-level object grounding, object detection, cropping, tracing, pixel difference between screenshots, long-screenshot OCR, foreground extraction, dominant color extraction, HTML screenshots into DeepSeek Harness for text-based models. DSH Vision Toolkit comes with an integrated free Gemini 3.7 Flash vision engine(300 images/day/machine without registration) or allow you to use your custom OpenAI or Anthropic compatible model.
Why a vibe-coder should care
For those who need to show the agent what they see on their screens like reconstruct a UI design based on a reference screenshot, compare different versions of a UI, or extract icons from a screenshot, DSH Vision Toolkit enables it in DeepSeek Harness without paying for the vision model.
▌ More finds