LLM Gateway in plain English.

An LLM gateway is a service or library that gives your application one API for many models and providers, and takes care of the operational mess in between: routing each request to the right model, falling back when a provider is down or rate-limited, managing keys and spend limits, caching repeated prompts, and logging every call for cost and quality analysis. OpenRouter is the best-known hosted gateway, exposing hundreds of models behind one key; teams also self-host gateways or use observability platforms that include the function. Gateways matter more as products use several models — a cheap one for classification, a frontier one for hard reasoning — and as pricing changes weekly. The trade-off is one more hop of latency and a new dependency in the critical path.