Choosing which AI models your team can use

An owner or admin decides which of the AI models available on your Profusia the workspace may use, under AI access in the workspace settings. Tick the models you approve; the rest are struck out for everyone. The list only ever narrows: it cannot add a model the deployment does not offer. The same list can be narrowed again for a group, and again for one person. Select any number of groups and people at once and add or remove models for all of them in one move. A person's list can only be a subset of their groups', which can only be a subset of the workspace's — and a person who has been given nothing gets nothing, rather than everything. Some models can be asked to think harder. For those, you can hold the workspace, a group or a person to a range of effort — say, medium at most for everyday work — and the range narrows down the same three levels. Each model shows its published price per million tokens, a modelled cost for one everyday answer, and what it has actually cost this workspace this month. Whether members see the cost of their own answers is your call, for the whole workspace, a group or a person. When it is shown, every answer in the conversation drawer carries a line saying which model answered and roughly what it cost. When a person has more than one approved model, the drawer lets them pick one, and an effort where the model allows it. A pick outside what they are allowed is refused in a sentence, never quietly swapped. A model too small to read what the question needs, or one on a provider tier that may train on what it is sent where the workspace has not accepted that, is refused too. The page also offers short, computed recommendations — the cheapest approved model that can read a whole document, where spend is concentrating, where an effort ceiling would save the most — and a market view drawn from OpenRouter's public model list, named as theirs, with links out to independent comparisons. None of it is a benchmark and none of it is a quote.