Vision Fallback
lelabdev
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.