What the page defines
An AI gateway, per nRouter's page, is the architectural control layer between user-facing applications and upstream AI model providers. Unlike a simple proxy, it handles multi-modality routing, dynamic provider failover, prompt-injection defense, PII redaction, token spend ledgers, and exact list-price settlement. The centralized controls: model routing and provider failover; guardrails and PII protection; budgets and spend limits.
The request path
Traffic passes authentication, guardrails, and a budget check before reaching an eligible provider. When a failure is retryable, the gateway retries against an allowed fallback provider or model, subject to capability and budget checks — and the request, response, latency, and cost are tracked through the whole path.
One entry point, many modalities
Text and multimodal generation, images, audio, speech and video jobs, embeddings, and model discovery share one governed entry point, for every capability the selected provider and model support. The connection is an OpenAI-compatible API: base URL https://api.nrouter.ai/v1, the key in NROUTER_API_KEY, and the same setup works from the OpenAI SDK, LangChain, LlamaIndex, the Vercel AI SDK, and CrewAI.
The page also separates the terms cleanly: an API gateway protects HTTP services but usually lacks model capability and token-cost context; a model router only chooses a model — routing is one part of a gateway that also governs credentials, safety, spend, and evidence.
Full page: https://nrouter.ai/product/enterprise-llm-gateway