DeepSee adds vision and smart model routing to DeepSeek Harness
DeepSee solves the problem of integrating visual context into text‑only AI coding sessions by routing tasks between DeepSeek Flash and Pro models while using Gemini for image understanding. It works as a plugin that adds a read_image bridge, auto and customizable routing modes, and a visual check gate that iterates until the UI passes. It is aimed at developers using DeepSeek Harness who need to preview UI changes and leverage vision without leaving their workflow. Compared to other plugins, it offers free Gemini key rotation, parallel subtask execution, and built‑in visual validation.
View on GitHub →chang416/deepsee