Built a VS extension to give vision capabilities to DeepSeek on Copilot

I built a VS Code extension that solves a problem I've been frustrated with: text-only models are amazing, but they can't see images. You know that moment? You're using DeepSeek because it's blazingly fast and cheap, but you paste a screenshot into Copilot Chat and it just... doesn't work. Or you have to switch to a slower, more expensive vision model just to understand one diagram. What if you could keep your favorite text model and just give it the power to see? The Extension OpenRouter Vision Proxy does exactly that: You pick any text-only model from OpenRouter (DeepSeek, Llama, Mistral, Qwen, whatever) You pick a cheap vision model (default: Google Gemini 2.5 Flash Lite) Drop an image into Copilot Chat The extension silently routes the image through the vision model, gets a description, feeds it to your text model, and streams the response back No new tools to learn. No new sidebar. No extra processes running. Just... Copilot Chat, but better. Features At a Glance ✅ Image support for any text-only model ✅ Choose your models from two dropdowns ✅ Smart retries (0–5 attempts) ✅ Tool calling & agent workflows ✅ Secure by default (OS keychain) ✅ Zero runtime dependencies ✅ MIT open-source Install Grab it from the VS Code Marketplace Cmd+Shift+P → "Set OpenRouter API Key" Pick your text model, pick your vision model Paste an image into Copilot Chat Watch it work Curious? Try it and let me know what breaks. I'm running it daily with DeepSeek + Gemini, and it's been rock solid. Would love to hear if there are models/workflows that don't work well.