On setup it recommends a local model based on the hardware Otis is running on, downloads it and runs it through llama.cpp for you. Ollama, LM Studio and Nvidia PAIR are supported too.
Hope it makes it easy to try and use open-weight models for some of you and all feedback is welcome!