Thanks — good question. You're asking whether the

AgentSession

in LiveKit is associated with the inference gateway. Yes. An

AgentSession

uses LiveKit Inference as the unified gateway for the voice-AI pipeline (STT → LLM → TTS). The session’s

inference

configuration routes model calls through LiveKit Inference so you can specify provider models (for example,

deepgram/*

for STT,

openai/*

for LLM, or

cartesia/*

for TTS) without providing each provider’s API keys yourself. What this gives you: • A single interface to configure STT, LLM, and TTS models for a session (the

inference

module). • Consolidated routing, latency optimization via global co-location, and unified billing handled by LiveKit Inference. • Simpler orchestration of the voice-AI pipeline inside the

AgentSession

. If you’d like, I can: • Show a small example

AgentSession

configuration for the specific providers/models you plan to use. • Explain how billing or concurrency is handled for multiple providers in a session. Which would be most helpful? Sources: OpenAI LLM models | LiveKit Documentation | Simple Call Answering Agent | LiveKit Documentation | Agent session | LiveKit Documentation | Introducing LiveKit Inference: A unified model interface for voice AI | Liv...