Any estimates as to performance improvements (reduced latency) when leveraging LK inference especially when using avatars?

Is your agent hosted on LiveKit? you'll get latency improvements for STT-LLM-TTS, which are upstream of the avatar call. We don't directly support avatars in our inference service. But your end to end latency should improve because the inference service improves latency upstream of the avatar call. For self hosted agents, see my response here. https://livekit-users.slack.com/archives/C09J2AYGN0M/p1759468506841849?thread_ts=1759355716.866269&cid=C09J2AYGN0M