Hybrid Voice Agent Sample

An iOS voice agent that runs the whole conversation on the device — speech to text, a small language model, and text to speech — and can hand the "brain" over to a cloud model when you want it to. Speech recognition and synthesis stay on the device on both paths, so only the intelligence swaps, and the two share one transcript.

Built with the Switchboard SDK . The on-device voice pipeline comes from EdgeSpeech .

Status: early. The on-device speech pipeline runs, both brains sit behind one interface, and you can switch between them mid-conversation.

We originally built this example as part of a buildathon. This Loom video provides a brief intro: https://www.loom.com/share/a6af3e55e93246b79cf71a28d54e96cd

And the video below provides a more thorough overview of the functionality.