in livekit room avatar will be considered as remote participant or local participant ?

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

talking from client side (Backend side)

Thanks — you clarified you mean both the client side and backend. I’m responding to whether an avatar in a LiveKit room is considered a remote or local participant from both perspectives. Short answer: the avatar is a remote participant on both client and backend. Why: • The avatar runs as a separate participant (an “avatar worker”) that joins the room independently and publishes audio/video tracks. • It is distinct from the original agent participant; the avatar worker publishes on behalf of the agent but is still a separate (remote) participant. How to identify the avatar programmatically: • Look for participants with

kind

set to

agent

. • Use the

lk.publish_on_behalf

attribute: ◦ The original agent participant has

lk.publish_on_behalf = null

. ◦ The avatar worker has

lk.publish_on_behalf = <agent_identity>

indicating it publishes on behalf of the agent. Example detection logic (conceptual; adapt to your SDK/language): • On the client (or any backend using the LiveKit room participant list): a. Find the original agent: • participant.kind === 'agent' && participant.attributes['lk.publish_on_behalf'] === null a. Find the avatar worker: • participant.kind === 'agent' && participant.attributes['lk.publish_on_behalf'] === agent.identity If you want, I can: • Provide a concrete code snippet for your SDK/language (e.g., JavaScript/TypeScript, Go, or Python). • Show how to filter avatar participants in your UI or backend logic (e.g., routing media, applying permissions, or analytics). Which SDK/language are you using on the client and backend?

ha use published track event and print the track and participant identity in python SDK

Thanks — I’ll show how to listen for published tracks with the LiveKit Python SDK and print the track info plus the participant identity. This works for regular participants and avatar workers (which appear as remote participants). 1) Listening for newly published tracks

from livekit import rtc

# assume `room` is your connected Room instance

@room.on("track_published")
def on_track_published(publication: rtc.RemoteTrackPublication, participant: rtc.RemoteParticipant):
    # publication.track may be None until the track is available
    track = publication.track
    track_sid = track.sid if track else publication.sid
    track_kind = track.kind if track else publication.kind
    print(f"Track published: sid={track_sid}, kind={track_kind}, participant_identity={participant.identity}")
    # if you need to detect avatar workers:
    publish_on_behalf = participant.attributes.get("lk.publish_on_behalf")
    print(f"  lk.publish_on_behalf={publish_on_behalf}")

2) Inspecting already-published tracks after joining

for participant in room.remote_participants.values():
    for pub in participant.track_publications.values():
        if pub.track:
            print(f"Existing track: sid={pub.track.sid}, kind={pub.track.kind}, participant_identity={participant.identity}")
        else:
            # publication exists but track object not yet available
            print(f"Existing publication: sid={pub.sid}, kind={pub.kind}, participant_identity={participant.identity}")
        # detect avatar worker vs original agent:
        publish_on_behalf = participant.attributes.get("lk.publish_on_behalf")
        print(f"  lk.publish_on_behalf={publish_on_behalf}")

Notes • Use

participant.attributes['lk.publish_on_behalf']

to tell avatar workers apart from the original agent: ◦ Original agent:

lk.publish_on_behalf

is

None

. ◦ Avatar worker:

lk.publish_on_behalf

equals the original agent’s identity. •

publication.track

can be None briefly; use `publication.sid`/`publication.kind` until

track

is available. • The

track_published

handler receives both the publication and the participant, so you can log identity immediately when the event fires. If you want, tell me: • which LiveKit Python SDK version you’re using, and • whether you want the snippet to also subscribe to the track or save/forward its frames — and I’ll provide a tailored example. Sources: client-sdk-js | livekit/livekit_composite | DeepWiki | An open source stack for real-time multimodal AI | Job lifecycle | LiveKit Documentation