The same agent answers in the embedded widget, by voice, in Slack, in Teams, through Claude, and over email. It reads the screenshot and the policy, validates the numbers against BigQuery, and stops at the decision. The widget is configurable, point it at any agent, any surface.