1min.

1MIN · AI & AGENTS

Vercel routes self-hosted models through its existing gateway

Vercel AI Gateway’s Private Inference connects a team’s own compatible model endpoint to the same authentication, request formats and observability used for catalog models. Private model IDs are visible only to the owning team. The endpoint must still be publicly reachable over HTTPS, so private here describes team-scoped access through the gateway rather than an isolated network connection.

Vercel · Nov 16, 2026

Why it matters

A common inference interface can simplify provider switching, but routing, fallback and billing behavior remain part of the application’s design.

More context

Register an endpoint URL, API format, credentials and model through the dashboard or CLI 65.16.0+. Try a test model and inspect logs and fallback behavior. Official docs describe a default $0.20-per-1,000-request gateway fee without a qualifying Vercel-provided fallback, plus direct upstream costs.

Keep in mind

Released integration. A qualifying catalog fallback can remove the gateway fee but can also send a failed request to another provider. Private-model observability does not show the provider’s direct inference bill. Team scoping does not itself prove zero retention by the endpoint.

Original read

More engineering ideas, one minute at a time.

Explore more stories