The LLM has no memory. What you actually feed it is the entire conversation (truncated up to its input limit) every time you respond. But you are bound to the

The LLM has no memory. What you actually feed it is the entire conversation (truncated up to its input limit) every time you respond. But you are bound to the interface, especially if you are not running a local LLM like Llama. Different models have different prompting syntax.