This is the case with all LLMs as far as I know.

With the chatGPT api you just send everything up to that point + the new input to get the new output.

I think the benefit for the service is that it’s stateless. They just have requests in and out and don’t have to worry about anything else.