ElevenLabs Voice Engine Wraps Any Chat Agent in Voice

The Breakthrough

ElevenLabs Voice Engine provides a first-class primitive that wraps any existing chat agent with voice capabilities, including speech-to-text via Scribe, text-to-speech via V3, emotion-context-aware turn-taking, semantic barge-in detection, and support for thousands of voices and languages, all without altering the agent's core logic.

Integration Techniques

Developers create a Voice Engine instance in the server SDK, attach a wrapper to their existing chat agent, and start a session loop that proxies audio inputs and outputs. The client SDK requires three lines of code to embed a voice widget on a site, which also enables telephony and CEAS integrations out of the box. Pre-built UI components follow Shadcn and Vercel styles; developers point a coding agent at their codebase and issue one prompt to analyze, wrap, and deploy the agent as voice-enabled.

// Server SDK example structure
create client
create voice_engine
attach to existing chat agent
new session -> proxy loop

Tool calling passes through unchanged since the existing agent handles it on the backend; client-side tools manipulate the DOM, server-side tools run normally, and future updates will proxy additional calls to the wrapped agent.

Demo and Paradigms

Luke Harries demonstrated converting a local generic chat support agent to voice: the agent responded to 'hello how are you' in text, then the wrapper added voice proxying in the background. Voice Engine serves as a higher abstraction bundle over raw TTS/STT, contrasting full agent platforms; teams choose wrappers for existing agents or out-of-the-box platforms for new builds. Harries predicts chat agents will add voice layers or become obsolete, unlocking Zoom calls, phone lines, and ambient interfaces.