Regional inference now available on AI Gateway
AI Gateway can pin inference and retained provider data to the US or EU, giving agent builders one residency control with per-response verification across supported providers.
AI Gateway now accepts **inferenceRegion** to pin supported requests to the **US or EU**. The provider runs inference there, stores any retained data there, and the response identifies the serving region.
Use the field for agents handling residency-sensitive code or documents, and check model support through the model list or its regions array. Unsupported pinned requests fail instead of silently leaving the selected region.
AI Gateway now accepts **inferenceRegion** to pin supported requests to the **US or EU**. The provider runs inference there, stores any retained data there, and the response identifies the serving region. Use the field for agents handling residency-sensitive code or documents, and check model support through the model list or its regions array. Unsupported pinned requests fail instead of silently leaving the selected region. Residency is opt-in: requests without the field route globally. Regional rates are set by providers and are often **about 10% higher**, while model availability varies by region.
This makes residency an explicit, fail-closed routing constraint rather than an assumption about a provider or model. It narrows gateway selection for sensitive workloads to region-supported combinations and adds a visible tradeoff among compliance, availability, and roughly 10% higher provider pricing, while leaving global routing as the default.