Off Grid AI Desktop Lets Mac Users Keep Several Local AI Models Loaded at Once

Off Grid AI Desktop can keep multiple chat models resident in memory on an Apple silicon Mac, so requests naming a kept model are served immediately without unloading and reloading. The feature, called Keep loaded, is part of the app's gateway and is available in the free stable release 0.0.55 or later. The app plans memory against about 75% of unified memory, minus a 1 GB margin and the memory used by its own chat model, and shows a bar indicating whether the chosen set fits. Users select downloaded chat models in the Gateway section and click Keep loaded to start a llama.cpp model router, with the set reloading automatically on app launch. Untick models and save to free their memory.