AI gateway incidents usually start with a bad assumption about where trust begins and ends. This course shows how a self-hosted Node proxy can sit in front of multiple model providers, enforce caller identity, scope, routing, quotas, and safety controls, and still leave a clear record of what it can and cannot stop. It is aimed at platform engineers who need one central gateway for internal and customer-facing traffic, with controls that hold up under real failure modes and cross-provider complexity.
The course covers gateway trust boundaries, caller authentication, model allowlists, provider credential handling, transport and egress limits, shared rate limits, token and cost quotas, safety filters, residency-aware routing, tenant-isolated caching, telemetry for drift and abuse, and a regression suite for control coverage. It also maps controls to threats and clarifies residual risk outside the gateway.