OpenRouter Vision Proxy for Copilot Chat
Give any text-only model the power to see images — right inside GitHub Copilot Chat.
Love the model you've picked but wish it could read a screenshot? This extension drops an image-aware model into Copilot Chat that routes your images through a vision model on OpenRouter, then feeds the description to your chosen text-only model — all with your own API key.
Why this extension?
- Don't switch models — power them up. Paste a screenshot, diagram, or UI mockup into the chat you already use. No new sidebar, no new tool to learn.
- Use any text-only model on OpenRouter. DeepSeek, Llama, Mistral, Qwen — if it can't see images, this extension makes it see.
- Vision on a text-only model. Drop an image into chat and it's transparently described by a cheap vision model, then answered by your text model.
- BYOK — pay OpenRouter directly. Your API key, your bill, your rate limits. Stored in the OS keychain, never on disk.
- No Copilot subscription for inference. Copilot Chat is just the interface; all model calls go straight to OpenRouter.
Features
Image support for any text-only model
Paste screenshots, error messages, charts, photos, or UI designs into Copilot Chat. The extension hands the image to a vision model (default google/gemini-2.5-flash-lite), gets a description, weaves it into your prompt, and streams a smart answer back from your chosen text model.
Choose your models from OpenRouter
Two simple dropdowns pull the live OpenRouter model list — pick your text model and your vision model in seconds. Change them anytime.
Reasoning effort control
Tune how much your reasoning-capable text model thinks (none / low / medium / high) right from settings — handy for balancing speed and depth.
Smart retries
Transient hiccup? The extension automatically retries your request (0–5 attempts) on network, timeout, or rate-limit errors so you get your answer.
Because this plugs into Copilot's provider API, tool calling and agent workflows keep working alongside image support.
Secure by default
Your OpenRouter API key lives in VS Code's SecretStorage (OS keychain on macOS / Windows / Linux). Never in settings.json, never in your Git history.
Zero runtime dependencies
Pure VS Code API + Node.js built-ins. No Python, no Docker, no local proxy server to babysit.
Getting Started
Prerequisites
- VS Code 1.116 or later.
- GitHub Copilot Chat (for the chat interface).
- An OpenRouter API key.
Installation
- Install the extension from the VS Code Marketplace.
- Open the Command Palette (
Ctrl+Shift+P / Cmd+Shift+P).
- Run OpenRouter Vision Proxy: Set OpenRouter API Key and paste your key.
- Run OpenRouter Vision Proxy: Choose Text Model to pick your text model.
- Run OpenRouter Vision Proxy: Choose Vision Model to pick your vision model.
- In Copilot Chat, select the OpenRouter Vision model and paste an image. Done!
Settings
| Setting |
Default |
Description |
openrouter-vision-proxy.textModel |
deepseek/deepseek-v4-flash-0731 |
Text model that generates answers |
openrouter-vision-proxy.visionModel |
google/gemini-2.5-flash-lite |
Vision model that describes images |
openrouter-vision-proxy.reasoningEffort |
medium |
Reasoning effort (none/low/medium/high) |
openrouter-vision-proxy.maxRetries |
0 |
Retries on transient failures (0–5) |
openrouter-vision-proxy.endpointUrl |
https://openrouter.ai/api/v1/chat/completions |
OpenRouter endpoint |
openrouter-vision-proxy.visionPrompt |
(built-in) |
Prompt sent to the vision model |
openrouter-vision-proxy.visionTimeoutMs |
30000 |
Vision request timeout (ms) |
Compared to alternatives
|
This extension |
Native vision model |
Local proxy (e.g. LiteLLM) |
| Works inside Copilot Chat |
✅ |
✅ |
⚠️ |
| Use your favorite text-only model |
✅ |
❌ needs vision model |
✅ |
| Cheaper vision routing |
✅ |
❌ pays full model |
⚠️ |
| No extra process to run |
✅ |
✅ |
❌ |
| One-click install |
✅ |
✅ |
❌ |
| API key in OS keychain |
✅ |
✅ |
⚠️ varies |
Keywords
openrouter · vision · image · copilot chat · github copilot · multimodal · screenshot · image understanding · text-only model · byok · language model · chat
License
MIT
