Licensed to be used in conjunction with basebox, only.
// admin
Model permissions
Overview
Not every model has to be available to every person – for example an expensive or slow model only for an analysis team. Today basebox controls model use via organisation and app, not via an assignment of models to groups. This page shows how to use that to restrict access to models deliberately.
What it is for
- Cost: a large model consumes more tokens per answer. On basebox Server it also ties up GPU capacity.
- Suitability: a reasoning model is right for analyses, unnecessarily slow for short answers.
- Rollout: a new model should be tested by a pilot team before everyone gets it.
How basebox assigns models
| Level | Setting | Effect |
|---|---|---|
| Organisation | Model selection | The default model for all chats and for apps set to "Use default" |
| App | App editor → Advanced fields → Model | A fixed model for exactly this app, deviating from the default |
| User | – | In the chat, users choose no model but only control the reasoning effort. In their own apps they can choose via the Model field – a restriction to a group cannot be bypassed that way, though |
Which models appear in the app selection at all is determined by the organisation – see Enabling models.
Where to find the setting
- Default model: your organisation's administration area (Model selection)
- Model per app: "All apps" → Edit app → Advanced fields tab → Model field
Step by step: restricting a model to a group
- Choose as the organisation's default model the model everyone should use – usually the cost-efficient one.
- Create an app (or use an existing one) and, under Advanced fields → Model, set the model that only the group should get. Adjust the thinking effort if needed.
- Share the app via App sharing only with the desired group – with READ.
- Leave all other apps at "Use default".
Result: the restricted model answers only in this app, and only group members see it. The normal chat and all other apps continue to use the default model.
The reverse – a cheap model for a bulk task – works the same way: an organisation-wide published app with a fixed small model.
Notes
Note
- API keys can use all models listed under
/v1/modelsfor the organisation; there is no restriction per key. See API permissions. - Reasoning costs tokens. Besides the model, an app's thinking effort strongly affects consumption. Set it deliberately per app.
- Consumption per app is shown in the Dashboard (tokens, users, amount) – so you can tell which app ties up which budget.
- Limits per organisation cap total consumption regardless of model. See Settings.
- On basebox Server, model selection depends on which endpoints the Platform Operator has connected – see Models & inference.
Frequently asked questions
Can I assign a model directly to a group? Not as a dedicated setting. The way is an app with a fixed model, shared with the group.
Can users choose a different model in the chat? No. They choose the reasoning effort; the model is determined by the organisation or the app.
What happens if I remove a model from the organisation that an app uses as fixed model? The app can no longer answer until you assign it an available model or set it to "Use default".
Can I see which model is set in someone else's app? Yes, as administrator you can edit every app and view the Model field.
Need help? Contact support