AI Gateway
Every model request an agent makes passes through AI Gateway. Configure the model families your organization approves once, and agents then request a capability tier — Economy, Smart, or Premium — rather than naming a model.
That indirection is the point. When a better model arrives, an administrator remaps the tier and every agent picks it up. No agent is edited, and no model name is hard-coded in an application.
Configure and verify
Choose the families
Enable each approved family in Gateway Settings and designate one as the default. The default serves any request that doesn't name a family.
Audit before you deploy
Open Model Catalog and confirm each provider you rely on shows Enabled before agents depend on it.
At a glance
What it controls
Which model families, providers, and models agents can reach
Who configures it
An administrator with Control Center access
Scope
Family availability, credentials, and tier mappings are configured per project
Where usage appears
Credit Usage for managed-key spend
Where a served request shows
Runs & Traces shows the model and provider
Where changes are recorded
Where requests run
The provider you authenticate decides which cloud account and region serves the inference, and a bank-owned deployment can keep the whole runtime inside its own perimeter. See Security FAQ for deployment architecture, data residency, and what the platform logs.
Related
Last updated