AIProxySwift now supports the generally available OpenAI Realtime API. Existing Realtime integrations can migrate to the GA event and configuration shapes, while new apps can start with typed Swift messages, protected WebSocket connections, and audio helpers built for Apple platforms.

GA events and configuration

The update adds the GA session configuration, response creation, and conversation item types while preserving compatibility helpers for existing callers. It also expands the events your app can observe, including response lifecycle events, audio and transcript deltas, speech detection, conversation metadata, errors, and usage information.

Realtime tool configuration includes built-in web search, so a voice experience can decide when current information is needed and return an audio response with its transcript.

Clearer two-way voice on iOS

Our audio path can now use Apple's voice-processing I/O for echo cancellation. Microphone capture and model playback share the voice-processing route, reducing the chance that the model hears its own response and interrupts itself. Apps that need more control can opt into manual echo cancellation through AudioController.

We also added interruption handling so playback can stop as soon as the user begins speaking, an important detail for natural back-and-forth voice interfaces.

A sample project to start from

The OpenAI Realtime starter project demonstrates the full iOS flow. The AIProxySwift README also includes a standalone example and GA migration notes.