Vision Plugins

Vision Fallback

lelabdev

3

Routes image analysis to a dedicated vision model when the main chat model doesn't support vision. Intercepts images before they reach the LLM, sends them to a configurable vision model (via LiteLLM), and replaces the image with a text description. Works with any text-only model like glm-5.2, deepseek, or llama3. Features auto API key detection from .env, hash-based cache, and dark mode UI.

Vision Sidecar

GreifMax

2

More tolerant vision_load + addition of a Vision Model. When a non vision-capable Main calls vision_load, it can add a specific query to send together with the image(s) to the vision model.