OpenCode plugin that proxies images through a vision-capable model, enabling image-incapable models to "see" pasted images.

Add this to opencode config (typically in: ~/.config/opencode/opencode.json)

{
  "$schema": "https://opencode.ai/config.json",
    "plugin": [
      "@sami7786/opencode-image-proxy@latest"
    ]
}

1. Checks if the current model is in the "imageIncapableModels" list as configured
2. If image-incapable: Automatically sends images to a vision-capable model for analysis, then replaces the image with a text description
3. If image-capable: Images pass through natively with no transformation

All model calls use OpenCode's existing authentication - no separate API keys needed.

Create ~/.config/opencode/opencode-image-proxy.json (or $XDG_CONFIG_HOME/opencode/opencode-image-proxy.json)

To see available models, run: opencode models

default config:

{
  "imageIncapableModels": [
    "zai-coding-plan/glm-4.5",
    "zai-coding-plan/glm-4.5-air",
    "zai-coding-plan/glm-4.5-flash",
    "zai-coding-plan/glm-4.6",
    "zai-coding-plan/glm-4.7",
    "zai-coding-plan/glm-4.7-flash",
    "zai-coding-plan/glm-5"
  ],
  "imageReaderModel": {
    "providerID": "opencode",
    "modelID": "kimi-k2.5-free"
  },
  "analysisPrompt": "The user has pasted an image into their chat. Describe what you see as if you are directly observing the image. Be thorough but concise. Include:\n- All visible elements (objects, text, UI elements, people, etc.)\n- Exact transcription of any text\n- The context and purpose of the image\n- Any relevant technical details\n\nDescribe it naturally, as if explaining to someone what you're looking at right now."
}

• imageIncapableModels — Models that you want to enable the proxy for.
• imageReaderModel — Any vision-capable model in your OpenCode setup.
• analysisPrompt — System prompt for image analysis.

[User pasted image: screenshot.png]
This is a terminal window showing a Node.js error. The error message reads:
"TypeError: Cannot read property 'map' of undefined" at line 42 in app.js.
The stack trace below shows the error originated in the UserList component...

• The configured imageReaderModel must be available and authenticated in OpenCode
• The vision model must support image input

npm install
npm run build

MIT