The chat messages API

Every LLM API call to a major provider has the same basic shape. You send a model name and a list of messages. Each message has a role: system (or developer) for your instructions, user for what the person typed, and assistant for the model's earlier replies.

You also send a maximum output length and sampling settings. You get back the reply, plus a usage count of the tokens used. The API is stateless, meaning it remembers nothing between calls.

So a multi-turn conversation is your code adding each reply to the list and sending all of it again. Read the usage field on every response and log it, because that is your real cost. Learn one official SDK (the provider's own library) properly before adding a framework. Then you know what a request looks like when something breaks.