Role assignments

You can control which provider or group handles each role at any time.

• Namespace level: open the namespace settings and choose LLM Assignments. Each role row shows the current provider or group, whether it is an override or inherited from the global baseline, and a "Clear override" action. Only capability-compatible providers and groups are offered per role.
• Embedding: the namespace settings also have an Embedding Provider section, with the same full re-index warning described on The embedding index.

How roles resolve🔗

Roles resolve through provider groups. Your own providers in a group are preferred over the platform's. Any role your providers do not cover falls back to the platform's group members. This is why you can see platform models in your token spend even though you brought your own model. If you want a role on your model, check the group membership and the role's assignment.

The full list of roles and the capabilities each one needs is on the Capabilities page.