Skip to navigation

Models

Control which models employees reach, how requests route, and what they cost

Employees point their AI client at Gateway, so every model request passes through the controls you set under Models. Roll out models has the client setup.

ControlTabsPage
Which models employees can reachAll models, Blocklist, Provider keysModel access
How requests routeRouting policiesRouting
What gets scannedDLP, Security alertsModel guardrails
What requests may costBudget, under ManageBudget
Organization-wide settingsAdvanced (gated; use the Gateway console for now)Advanced

Order a model request passes through

  1. Access. Blocked models and vendors are removed. No routing policy can reach a blocked model.
  2. Guardrails. The prompt is screened and redacted, blocked, or recorded.
  3. Routing. The Group’s policy, or the organization default, picks a model and vendor.
  4. Budget. An exhausted budget refuses the request with a 402 before any provider is called.
  5. Response. The completion is screened, and the request is written to LLM calls.

A request that names a model skips only routing. A refusal reaches the employee as an ordinary API error.