Add PersonaPlex (NVIDIA) realtime plugin

Agents
1

Hi team, requesting a review for the PersonaPlex plugin PR.

What it does: Adds a new realtime model plugin for NVIDIA PersonaPlex — a full-duplex speech-to-speech conversational AI based on the Moshi architecture. It enables voice agents with persona control through text prompts and voice conditioning.

Key features:

  • Full-duplex audio streaming over WebSocket with Opus encoding (via sph)

  • 18 built-in voice presets

  • Automatic reconnection with exponential backoff

  • WSS/SSL support

  • Works as a drop-in RealtimeModel for AgentSession

Usage:

from livekit.plugins.personaplex import RealtimeModel

session = AgentSession(
llm=RealtimeModel(
base_url=“your-server:8998”,
voice=“NATF2”,
text_prompt=“You are a friendly assistant.”,
),
)

PR: Feat/personaplex plugin by milanperovic · Pull Request #4660 · livekit/agents · GitHub

The plugin has been tested with community members and several fixes have already been applied based on their feedback (sphn 0.2+ API compatibility, error recovery, session isolation). Would appreciate a review when you get a chance!

1 Like
2

Hi, let me ping the team as I know there has been a lot of interest about this & thanks for letting us know this has already been tested by community members. I see there is a recent review comment from Devin also, if you could please address that.

1 Like
3

hello! is this available already?

4

Looks like the PR is still open. But you can try it

1 Like
5

I see this topic has a lot of views, so to update, PersonaPlex support was added in LiveKit agents for Python 1.5.1, Release livekit-agents@1.5.1 · livekit/agents · GitHub. The docs are still under development and will be updated soon.

1 Like