Developer MaxiDonkey has released Edge-OpenAI-Realtime, a native VCL component that brings OpenAI’s Realtime API to Delphi applications using Microsoft Edge WebView2 and WebRTC.
Rather than treating voice as a series of speech-to-text and text-to-speech requests, the component enables true low-latency voice conversations with OpenAI’s realtime models. It encapsulates the entire WebRTC stack behind a Delphi component while exposing a clean event model for session management, audio streaming, and realtime interactions.
The project demonstrates:
- Live voice-to-voice AI conversations
- OpenAI Realtime API integration
- WebRTC audio and DataChannel communication
- Semantic VAD with interruption (barge-in) support
- Secure ephemeral authentication
- Reusable UI-free realtime architecture
- VCL component integration with WebView2
The realtime implementation is intentionally separated from the WebRTC transport layer through an abstract interface, allowing alternative WebRTC engines to be used in the future while preserving the same Delphi API. The repository also includes a complete VCL sample application demonstrating the component in action.
One item I would note is that FireMonkey does have a TWebBrowser component that supports Edge.
Check on GitHub to build low-latency voice AI applications with OpenAI’s Realtime API in Delphi.