A newly opened side chat shows an almost empty transcript. What does the model answering in it have to work from?

The principle — A side chat is seeded with the parent conversation as hidden reference context. The model can read it; the transcript does not render it. Only your prompts and the replies to them appear there.

Why the key is correct — The empty look is a display decision, not a description of what the model knows. Asking about something discussed twenty messages up the parent thread works with no preparation at all.

Why the others are wrong — The context is more than what you type, so re-pasting the discussion burns tokens and muddies the question. The history itself is seeded rather than a summary of it, so detail survives. And more than the closing message comes across, so early parts of the investigation are answerable too.

Remember this — An empty side-chat transcript still has the parent history behind it.

Sources — Cursor's Side chats documentation.