Vercel shipped several small developer-platform releases in the same window. Read separately, they are changelog items. Read together, they show an agent platform control plane taking shape.
AI Gateway now supports regional inference. A request can set an inference region for US or EU, and if no provider can serve that region, the request fails rather than silently running somewhere else. The response reports the region that served it, making residency a verifiable runtime property.
Vercel Connect added preset connectors for 100+ services. The connector presets pre-fill service configuration, then applications request scoped, short-lived tokens for attached services. That is a useful agent primitive: the app does not need to hold broad permanent credentials when a task only needs a narrow capability.
Vercel Sandbox now supports Sandbox.fork(). A fork starts from the source snapshot, inherits config and environment variables, and can override selected resources. That is exactly the shape agent workloads need when several branches, tenants, or experiments start from the same base environment.
Python Functions now support WebSockets for ASGI and WSGI applications, including FastAPI, Django, and Flask. That matters for real-time AI streaming, live chat, and collaborative sessions.
New Runtime Read
The common thread is control: where inference happens, which external systems a task can access, how an execution environment branches, and whether the runtime can keep a live bidirectional session open.
Agent-ready infrastructure is not one magical endpoint. It is a bundle of boring but consequential controls. Vercel’s releases make those controls visible as product surfaces.