Which model answers

At the bottom right of the message box stands the model this conversation uses. Click it and you can choose another one — Model for this chat, and only for this chat.

Who decides what is offered

Your workspace's administrator sets which models may be used at all. The picker offers that list, not everything that exists.

That is deliberate: which model a company's data may be shown to is a company decision, not a per-person one.

On your own network

An entry can say on your network. Then the model runs on a machine you control — a team server in the office, for instance — and the conversation does not leave the building.

Worth choosing when the question touches something you would rather not send anywhere. Worth knowing when it does not: everything else goes through Pirol's endpoint, which is where personal data in a message becomes a question you can answer before you send it.

Why models differ

A bigger model reasons better and costs more; a smaller one answers faster. For "add this customer" the difference is invisible. For "compare these three offers and tell me what is missing" it is not.

If an answer disappoints, switching the model for that chat and asking again is a fair first move.

What the agent keeps between models

Nothing changes about your data or what the agent may do — the model is the part that thinks, not the part that holds permissions. Your history, your memory and your approvals stay as they are.