Model availability
Decide which AI models your workspace, a team or one person is allowed to use.
Settings → Governance → Models is where the catalog gets narrowed. The page says what it does: "Choose which AI models are available, for the whole workspace, a team, or one person. Turning a model off hides it from every generator and stops new spend."
Two reasons admins come here: cost, because the difference between the cheapest and the most expensive model in the catalog is enormous, and consistency, because a workspace often wants everyone producing work with the same handful of models.
Opening the page needs the permission to view workspace activity and governance. Changing anything needs the separate permission to change workspace governance settings. Without the second one the page still opens, with a notice at the top — "You don't have permission to change these settings. This is your workspace's current configuration." — and every switch inert. See Roles and permissions.
Three scopes, and they only ever narrow
At the top of the page is a Scope selector with three settings.
Workspace
"The default for everyone in the workspace." This is the baseline everybody starts from.
Team
Pick a team and you're editing rules on top of the workspace default. The page says so: "Rules on top of the workspace default. Anything turned off for the workspace stays off here."
Person
The same, one level down — rules for one individual, on top of both their workspace's default and their team's.
Restrictions only subtract. You cannot grant back at a narrower scope what a wider one turned off. A model switched off for the workspace is off for every team and every person in it, and the row at the narrower scope says which level did it — "Off for the whole workspace." or "Off for this person's team." If you need to give one team access, turn the model on at the workspace and off for everyone else.
Rows also tell you where their current state came from: "Inherited" when it's just following the level above, and "Off for this scope" when the switch you're looking at is the one doing the work. "Reset" clears the rules set at the scope you're viewing, putting it back to whatever it inherits.
What turning a model off actually does
It disappears
Not dimmed, not marked unavailable — gone. It leaves the model pickers
wherever you generate, it stops being offered by @ mention, and the chat
can no longer reach it.
It stops new spend
The block isn't cosmetic. A request that names a disabled model is refused when it's submitted, so nothing runs and nothing is charged — the interface and the server agree.
Work already produced with a model isn't affected. Turning a model off stops new runs; it doesn't remove or hide the images, videos or chats it already made.
Restricted is not the same as plan-gated
Two different things make a model unusable, and only one of them is yours.
Needs a paid plan
Stays visible in the picker — dimmed and unselectable, with the plan it needs named. Nothing you do on this page changes that; upgrading does.
Turned off by governance
Isn't in the picker at all. If a model has vanished rather than gone grey, this page is why.
That distinction is the fastest way to answer "why can't I use this model?" from a teammate: if they can see it, it's the plan; if they can't find it, it's a restriction. Models covers it from their side.
Media and Chat are separate catalogs
The page has two tabs.
Media
The models that make images, video and audio. Each one shows what it costs per run in credits — or "Varies" for models priced by the size or length of what you feed them.
Chat
The models that power the conversation itself. Governed the same way, from the same scope selector, by the same people.
This page deliberately doesn't list the models. The catalog changes without a new version of the app — models are added, repriced and retired continuously — so any list written here would be wrong within weeks. The page itself is your live list, complete with each model's current cost.
Working through a long catalog
The media tab is grouped into sections by kind, and each section can be collapsed or expanded. Alongside the search box ("Search models or providers…") there are filters for category and provider, an availability filter offering All, Enabled and Off, and sorting by name, provider, cost or availability.
The practical way to lock a workspace down is top-down rather than model-by-model: use a section's menu to disable all in section, then switch back on the two or three you actually want. A counter at the top tells you how many of the catalog are enabled, so you can check the result at a glance before you leave the page.
Restricting models is a narrower tool than it looks. If nobody in the workspace should be generating video at all, turning the whole surface off is the cleaner change — see Feature availability.