Documentation/Routing and guardrails

Routing and guardrails

Configure model access, retries and request protection.

Models and deployments

Applications send a public model name; the gateway selects an available deployment. Multiple deployments may serve one public name. Upstream and public names may differ.

Route templates

Configure weights, retries and timeouts in route templates and bind them to the intended scope. Parameters must fit deployment protocols. Verify in the playground first.

Guardrails and logs

Guardrails block, redact or allow according to rules and execution stage. Inspect rules, responses and logs to distinguish policy decisions from provider failures.

Search full documentation: budgets, routing, Docker…