Large language models are stateless. The apps and fine-tuned models are doing prompt engineering on users' behalf. It's very much a thing for developers, with the goal of making it invisible for end users.