Each seat's model picker has two parts: the vendor's own models at the top, and All models — a catalogue of every model we can reach, from every maker, with its price. Choose a catalogue model and the chair holds it, whoever made it: the seat is renamed for its occupant, gets that maker's mark, and answers to its own @ handle.
What powers it
A catalogue model runs on Our Keys and draws down your credit at the model's list price. A seat holding a catalogue model needs an account with credit, whatever key mode the seat was in before. A model the catalogue lists without tool support is asked without tools, so it can still review but cannot run commands on the Mac.
The reviewer follows the lead
When you put a catalogue model in the Leader slot, the Reviewer slot is aligned to it: the same maker's next-older release that is no dearer, or failing that its cheapest sibling. You can change it afterwards.
A custom endpoint
Set Runs on to Custom Endpoint and the seat talks to any OpenAI-compatible API: DeepSeek, Together, Azure, a local Ollama or LM Studio, or your own server. You give the endpoint's URL, its key, the model name it expects and a display name. The key is that endpoint's own, nothing passes through us, and Our Keys and credit do not apply. On the Mac the call goes straight from the Mac to the endpoint. On the free plan a seat on your own key, a custom endpoint, your company's cloud or a CLI works for one 14-day trial, then needs Pro; Our Keys and Free Key have no such limit.
A model pick saves itself on the web and the phone; Save is only for a key or an endpoint.