is it possible to have a custom speaking order between one user and multiple agents? I want agents to speak after one another

Thanks for reaching out, Nathan. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

will this allow agents to speak directly after each other, or only after the user?

Hey there! In LiveKit, agents can be set up to chat one after the other without waiting for the user to jump in, depending on how it’s set up. Could you let me know if you’d like the agents to speak back-to-back on their own, or only when the user responds? That way, I can give you the best advice possible. Sources: Agent speech and audio | LiveKit Docs | How does end-of-utterance detection work in conversations?

I'd like the agents to speak back-to-back on their own

If you want your agents to chat back-to-back in LiveKit without waiting for user input, you can do that by manually controlling when they speak. Here’s a simple way to set it up: 1. Manual Turn Detection: Use

turn_detection="manual"

in your

AgentSession

so you decide the flow of conversation. 2. Disable Audio Input: Turn off audio input at the start so agents don’t pause waiting for the user. 3. Sequential Speech: Call

session.say()

for each message you want your agents to say, one right after another. Here’s a quick example to get you going:

class BackToBackSpeechAgent(Agent):
    async def on_enter(self):
        self.session.say("Hello, this is the first message.")
        self.session.say("This is the second message.")

async def entrypoint(ctx: JobContext):
    session = AgentSession(turn_detection="manual")
    session.input.set_audio_enabled(False)
    await session.start(agent=BackToBackSpeechAgent())
    await session.commit_user_turn()

This way, your agents can have a smooth back-to-back conversation without any pauses for user input. If you have any questions or want to tweak it further, just let me know—I’m here to help!

is there a way to make use of the turn detection model while still achieving this behavior

Great news! You can definitely use turn detection models while letting agents chat back-to-back, it just takes a bit of thoughtful setup. Here’s a friendly guide on how to make it work: 1. Turn Detection Models: LiveKit’s turn detection models are pretty smart—they can tell when someone’s finished speaking. You can use these to smoothly switch between the user and the agent. 2. Custom Logic for Agent Turns: These models mainly focus on user input, but you can add your own logic to let agents jump in one after another once the user’s turn wraps up. 3. Manual Overrides: Mixing turn detection with manual overrides gives your agents the freedom to keep the conversation flowing without waiting on the user every time. This way, you get the best of both worlds—using turn detection while keeping control over how your agents respond. If you want, I’m here to share more details or examples anytime!

give me an example of how you can use the turn detection model AND the manual overrides

I don't have the answer you're looking for. You could also try asking your question: • in one of the other Slack channels or • to https://deepwiki.com/livekit/livekit_composite which is trained on all LiveKit source code If you find the answer, please post it here to help others!