GitHub radar

DSH Vision Router: Free Vision for Text-Only AI Agents

DSH Vision Router is an open-source plugin for DeepSeek Harness that gives text-only AI agents the ability to see images — for free, without an API key, up to 300 images per day via Gemini 3.7 Flash.

01ysr666/dsh-vision-router 955JavaScript

DSH Vision Router adds 14 visual tools to text-only agents in DeepSeek Harness: image Q&A, object grounding with pixel coordinates, object detection, cropping, SVG tracing, pixel diff between screenshots, long-screenshot OCR, foreground extraction, dominant color analysis, and HTML screenshots. When an image is pasted in chat, the plugin automatically routes the request through a built-in free Gemini 3.7 Flash vision chain — no configuration, no Python required, and image turns work exactly like ordinary tool-calling turns.

Why a vibe-coder should care

If you want your agent to understand a UI screenshot, locate interface elements by pixel, or read text from a long webpage, this plugin makes it possible at no cost — without paying for a vision model subscription.

Open on GitHub