I don’t know whether just using an LLM is a problem. But in your case I would say the fact you used one and didn’t indicate you did. If you indicated the answer came from an LLM, then the trust in the answer could be weighted accordingly by each user.
That’s my opinion at any rate.
You could plug in a USB SSD or HDD and make sure the DB and other regularly written data goes there. That would pretty much remove the problem.
I would wonder how well it would perform. The limited memory and cpu power surely would make database access not great under even moderate load.