Information has to earn the interruption.
Your existing AI gathers evidence. OnMe owns the rules around it: where the result belongs, what it may affect, how much reasoning it earns, and whether you should see it now.
Define the permission
Choose what starts the check, what it can affect, how much work it can spend, when it expires, and where the result goes.
Let your agent gather
ChatGPT, Claude, Cursor, or another capable host gathers the evidence and sends the result to OnMe.
Get the decision
OnMe returns PURSUE, WATCH, or SKIP, plus the reason, supporting evidence, work cost, and next check.
Correct the memory
Accept it, change it, reverse it, or record what happened. Cleared work leaves active attention.