Governance & Controls
Model controls
Choose which models your organization can use, and see what a change affects before it applies
Model controls let organization admins decide which models people in the organization can use. You can block a whole provider or a single model, decide whether models added to Relevance AI later start on or off, and see which features a change affects before you confirm it.
Open model controls
Open organization settings
Open your organization settings and select Integrations & models.
Open the Models tab
Select the Models tab. The count beside it shows how many models are on, for example 17 of 19.
Choose which models are allowed
The dropdown at the top of the Models tab sets every model at once:
- Allow all turns every model on, including new ones.
- Block all turns every model off. Agents and Tools using them stop until a model is switched back on, and new models are off by default.
- Custom lets you choose provider by provider and model by model. Your saved choices come back whenever you switch to Custom.
Each provider, such as Anthropic or Google, has its own count and its own Allow all, Block all, or Custom dropdown. Expand a provider to switch individual models on or off with the toggle on each row.
If an organization API key that a model relies on is switched off, the model shows API key off and is blocked until the key is turned back on. See Organization- and project-level API keys for managing those keys.
Control new models
Turn on new models automatically decides what happens when Relevance AI adds a model:
- Allow all turns new models on for every provider.
- Block all keeps new models off until an admin turns them on.
- Custom lets each provider decide, using the Turn on new [provider] models automatically toggle inside that provider.
Organizations that only allow approved vendors can use Block all so models released later never turn on without review.
See what a change affects
Relevance AI runs some platform features, such as memory, web reading, and error explanations, on models it chooses. Each model lists the features it powers under Powers. A red pill marks a feature that your current settings have turned off.
When your settings affect any features, a banner at the top of the tab shows how many, and how many stop or work with limits. Expand it to see each feature and the models it needs.
Before a model turns off, a confirmation lists what it affects:
- Stops marks a platform feature that won’t run without the model.
- Works with limits marks a feature that keeps running with less capability.
- Uses [model] instead marks a built-in Agent that moves to another allowed model.
- Your agents and tools lists your own Agents and LLM tool steps on that model, which stop until you switch them to another model or turn the model back on.
Turning the model back on restores everything the confirmation listed.