Gerbil is a desktop application for locally running large language models and image generation models. It uses KoboldCpp (a llama.cpp fork) for LLM inference with support for CUDA, ROCm, Vulkan, and Metal acceleration. The app includes built-in presets for Flux, Chroma, Qwen Image, and Z-Image workflows, integrates with HuggingFace for model browsing and downloading GGUF files, and can launch SillyTavern and OpenWebUI for extended interaction capabilities.