An LLM gateway is a single controlled path between applications and model providers, enforcing an approved model catalogue and centralising credentials, routing, rate limiting, fallback, cost attribution and policy on every request.
What is an LLM gateway?
Without a gateway, each application holds its own provider keys, chooses its own models and reports its own usage, which makes model policy unenforceable and cost unattributable. A gateway consolidates that path so a model can be approved or withdrawn centrally, credentials never reach application code, requests can be routed by sensitivity as well as by capability, and every call carries the identity it was made for. It is also the natural enforcement point for cost controls and for provider fallback.