livekit-voice-agent

Build LiveKit voice agents with multi-agent handoffs by chaining specialized agents like IntroAgent and EscalationAgent. Function tools return new agent instances to transfer control, maintaining shared conversation context across the routing workflow.

Context preservation during a voice agent handoff works by retaining shared user data across agents via chat_ctx. This ensures the receiving agent has full conversational history without losing information when routing between specialized agents.

You can deploy LiveKit voice agents using Docker for production environments. The implementation provides production-grade templates, prewarming, testing scaffolds, and observability hooks to support scalable and reliable real-time voice workflows.

Escalate a LiveKit voice conversation to a human by implementing an EscalationAgent within a multi-agent workflow. The bidirectional and escalation handoff patterns route the conversation dynamically while preserving the existing chat context.

You need to configure STT and TTS providers alongside LiveKit and OpenAI credentials in your environment file. These external services are required to enable real-time speech-to-text and text-to-speech processing for the voice agent.

Supported multi-agent workflow patterns for LiveKit voice routing include hub-spoke, escalation, and bidirectional handoffs. These modular patterns allow you to route real-time conversations between specialized agents based on dynamic function tool calls.