Skip to main content
Models run where the data lives. By default, each request routes to the closest available region. Two HTTP headers override that: X-Region-Override pins a specific region, and X-World-Part-Override keeps traffic within a broader geographic zone.

Default behavior

When you make a request without any region headers, it routes to the closest available region. No configuration needed.

World-part constraints

Keep traffic within a geographic zone with X-World-Part-Override. The best available region within that zone is selected. Use this when you need to stay within a geography but do not need a specific data center.

World parts

A world part groups several regions under one value. Requests route to the best available region within that zone, which does not guarantee a specific country.
If your compliance requirement depends on a specific country, use X-Region-Override with the exact region. A world part routes to the best region within a zone and does not guarantee a specific country.

Compliance

Regional execution is not just about latency. In regulated industries it determines where data can exist and which models can process it. These constraints are respected per request. If you need a particular model hosted in a region where it is not currently available, contact us.

Next steps

Models by region

See which models are deployed in each region.

Context Router

How geography is one of four dimensions the layer adapts on.