What do you mean by “stay alive”? LLMs are statistical models that require a lot of number crunching in order to output a response. That’s not something you can just host without some major costs associated (Far more than any single Fediverse instance for sure, GPUs are not cheap to rent).