An organization decides which models it allows; an agent offers one or more of them, and users can pick per message. The Auto Router, in early access, chooses a model per message and starts in shadow mode, so you see what it would pick before it answers. A model that is down hands over only to an allowed model at the same or a lower price.
Several AI models per agent, without changing anything for those that use one.
When a chosen model is unavailable, the answer comes from another model the agent already offers, never a more expensive one, and the reply says which model answered.
For models that support it, each agent can be set to spend less reasoning on simple work. In our test on 30 real support questions, output per answer fell by more than half with no additional errors.
We will walk you through it with your own data and your own channels.