Vercel AI Gateway adds private inference
Vercel AI Gateway now lets teams route requests to self-hosted, fine-tuned, and prerelease model endpoints using a unified private slug identifier. The feature brings custom models into the same authentication, SDK, observability, and fallback pipelines used for public catalog providers while isolating access to designated Vercel teams.
Unifying private model endpoints with commercial API gateways eliminates the fragmented tooling and duplicate auth logic that slows down enterprise AI deployments. Giving self-hosted infrastructure first-class status alongside frontier models makes hybrid AI stacks substantially easier to build and monitor.
- –Single interface: Developers query private endpoints through standard OpenAI, Anthropic, or AI SDK interfaces simply by specifying the model as private/<slug>.
- –Automatic fallbacks: Teams running self-hosted inference clusters like vLLM can configure seamless fallbacks to commercial catalog models whenever local compute fails or saturates.
- –Agent integration: Coding agents connected to Vercel AI Gateway—including Cursor, Claude Code, and Codex—can tap proprietary internal models without custom client configurations.
- –Enterprise security: Private models are isolated to team credentials, omitted from public catalogs, and support Zero Data Retention settings per provider.
DISCOVERED
20m ago
2026-11-16
PUBLISHED
28m ago
2026-11-16
RELEVANCE
AUTHOR
vercel
