The context meter
Every model has a limited context window — a cap on how much conversation it can consider at once. The usage panel (open it from the small indicator near the message box, or from the Usage tab in your account menu) shows how full that window is for your current conversation.
To stay within that budget, Vega only ever sends your system context plus the most recent turns of a conversation to the model — older messages further back are not included in what the model sees. This only affects what's sent to the model each turn; your saved chat history is never touched or deleted by this. You can always scroll up and see everything you and Vega have said.
Compact
When a conversation gets long, "Compact" summarizes the older portion into a short recap and keeps that instead of the full text, freeing up room in the context window without losing the thread. Clicking it now asks you to confirm first, since it changes what the model will consider going forward. It never deletes or edits anything in your actual saved conversation — only the copy sent to the model changes.
Clear context
"Clear context" wipes what's queued up to send to the model for the rest of this session, so the next message starts fresh. Like Compact, this only affects the model's working context — your visible chat log is left exactly as it is.
Daily limits
Signed-out (anonymous) use is capped at a small daily token budget meant for trying Vega out — that one's real, and resets daily. Signing in raises the displayed limit substantially, and the usage panel still tracks your usage against it and shows when it resets.