Skip to main content

Configure writer and fact checker profiles

Open the Account AI administration controls with admin.ai.models.read to inspect profiles. Editing requires admin.ai.models.manage. Complete account setup first.

Inherited defaults and overrides​

An empty model selection follows the current tenant default. Empty reasoning follows that model's reasoning default. Without an explicit default, the supported default value applies, or otherwise the first supported reasoning value. If the user's grant excludes that value, the call fails. The application does not silently choose another value.

Inheritance is a reference, not a copy. A later default change affects the next agent call. An administrator can save explicit model, reasoning, or prompt values. Choose Reset to inherited defaults, then Save agent profiles, to remove overrides. Empty prompts use the built-in role prompt. Personal profile overrides are unavailable.

The fact checker profile is configurable. Creating it does not start automatic fact checking. The writer produces draft content. Its profile does not grant Send permission.

Contract and authorization​

AdminService GET/PUT agent-profiles stores agent.profiles in authorization.policies, with an empty tenant subject. Each profile has nullable modelId, reasoningLevel, and systemPrompt fields.

Saving requires an active model and provider in the tenant and supported explicit reasoning. The server rejects unknown payload fields. The audit event is agent.profiles.update.

resolveAgentProfile(identity, role) resolves inheritance for each call and checks the caller's actual model grant. A tenant override does not grant access or remove context limits. An unavailable or forbidden model returns an error without silent substitution. See AI governance.