I think using some kind of MLM(medium size language model) that just knows how to interact with a KB/DB system and translates it into english(language of choice) would be much more valuable to a mid-size company/enterprise company.

doesn't everything you send to them allow them to train on it? how would you even know if your data went into another model?

Unless you build some kind of test data management system and the LLm provider promises not to compete seems like your kind of in a sticky situation.

If you're building with LLMs right now, what's your approach to defensibility?