I would enjoy training a LLM on my aggregated command history and using that for auto completion, or maybe using an open source one trained on a larger set from the community, but I am very uncomfortable sending data about every command (as I’m writing it!) to any company.