It can be powered by any locally hosted model including Gemini, I've also tested with very small models like llama3.2 and it works well.