What happens, in full
Going into a meeting prepared matters, and a back-to-back calendar makes it hard. You rarely have an hour to research someone properly, and often you do not have five minutes.
So you ask O the way you would ask a person. Name who you are meeting, say when, and name the framework you want to prepare against. That is the whole instruction.
O starts by finding the person, then looks through your own network for everything already attached to them: conversations in WhatsApp, threads in email, calendar events you have shared, notes you took at the time. In parallel it searches the public web for what is current, their role, where they have worked, and what they are active in now.
When it needs more context to be useful it says so and goes back for it, rather than filling the gap with a guess.
What comes back
A brief, not a list of links. Who they are publicly. What the two of you already share, read out of your own history, including when you last spoke and where you met in person. And a way into the conversation, shaped by the framework you asked for.
The second half is the part a research tool cannot reach. Anyone can read the public web. Only a system that also holds your side of the relationship can tell you this is not a cold meeting, and show you the five dated moments that prove it.
Why it can do this
Because the relationship history lives in one place. The conversations, the notes and the meetings are already there, so O can go back through them instead of asking you to remember. Five minutes before a call you ask, you let it run, and you arrive knowing exactly where you left off.