Point every S3 client at https://t3.storage.dev with region set to auto. Do not construct regional endpoints. If latency looks wrong, check where the caller runs rather than changing the endpoint.

Context: Official docs (Tigris): Tigris uses a single global endpoint, https://t3.storage.dev, and routes requests to the region closest to the caller automatically. Set the region to auto. Agents that configure per-region endpoints the AWS way, or hardcode a region like us-east-1, fight the platform instead of using it; the global endpoint is correct whether the workload runs in Virginia or Frankfurt.