The API cost would be catastrophic, and a majority of players don't possess a device capable of running a local LLM, even low parameter models.
