The Vercel platform opens the door to voice agents.

Vercel adds voice capabilities to AI Gateway, enabling real-time audio with OpenAI and xAI models like gpt-realtime-2 via its new AI SDK 7 features.

On AI Gateway, Vercel adds voice to the modalities already supported by its platform. Developers can now integrate real-time audio sessions, speech synthesis, and transcription with AI SDK 7, via `useRealtime`, `generateSpeech`, and `transcribe`.

The first available models come from OpenAI and xAI, including `gpt-realtime-2`, `grok-voice-think-fast-1.0`, `grok-tts`, and `grok-stt`.

The real-time mode allows an application to receive and produce audio directly, without separately chaining transcription, text model, and output voice. Calls go through AI Gateway with the same building blocks as other models: provider routing, observability, spend control, and external keys.

On the browser side, Vercel plans for a short, server-generated token to avoid exposing the API key to the client. The functions remain in beta.