Hi. Ever since upgrading from Livekit Agents 1.5.1. (Stopped workign at 1.5.2) Running in console mode has stopped working for us. Has anyone else had this? Stuck at "job runner initialized" when running "uv run python src/agent.py console". No on 1.5.5 and still have the same issue.

Hi — we're on the Ship plan (20-session cap) and seeing a pod-sizing / autoscaler issue that's killing sessions well under our plan cap. Project:

commandiq

(subdomain

commandiq-6ek744bb

) Agent:

CA_YHsMYA66ykvG

Pod:

deployment-p-6c3dvi560ji-ca-yhsmya66ykvg-699d9fcc59-tblcq

Timestamp of incident: 2026-04-22 130931 UTC (epoch 1776877771) What we ran: 5 concurrent agent sessions, each ~10 min, started within seconds of each other. Our per-session memory overhead is small (~30-40 MB). What we observed in our own agent logs: - All 5 sessions were dispatched to the same single pod (only one hostname in the log throughout) - Pid=1 supervisor logged

process memory usage is high

at 530 MB (warnMB=500) - ~2 seconds later, all 5 sessions reported

room_disconnected

simultaneously (session durations 131-137s) - No "registered worker" / second hostname appeared — i.e. a second pod was never provisioned What we expected: With a 20-session cap, we assumed LiveKit's scheduler would start spreading sessions across multiple pods before single-pod memory hit the SDK's warning threshold. Questions: 1. Is the scheduler supposed to provision additional pods before a single pod hits the memory-warning threshold? If yes, why didn't it here? 2. What's the documented "sessions-per-pod" target on Ship-tier pods? Is there a way for us to see or influence this? 3. Are Ship pods significantly larger than Build pods memory-wise, or is the plan cap raised without a proportional pod-size increase? Happy to share full agent logs, Playwright reproduction, or hop on a call. Our 3-concurrent runs are clean; 5-concurrent is 0/5 because of this pod-packing behavior.

This message was deleted.

This message was deleted.

This message was deleted.

how to track realtime metrics if the on_metrics_collected event is depricated

This message was deleted.

This message was deleted.

is realtime 2 available alr in livekit?

does this work for anybody? https://www.livekit.fm/

Hi, is anyone else experiencing issues with inbound calls in LiveKit?

Hi, is anyone else experiencing issues with automatically egress completed? I’ve faced this issue continuously several days.

Hello, can someone help in calling with ai , i recently upgraded my project in which i upgraded the livekit from 137.0.2 version to 144.0.0. in previous version calling was working fine but when i upgraded the project from then call is connecting with the ai connection is establisng but ai not speaking and microphone not enabling whereas sometimes it is working like if it worked in the first time then in the next call it won't work like this is happening, before establishing the connection i'm playing the ringtone using expo-audio, microphone permission is granted properly and i'm using physical device(iphone 16pro)

Hi, I don't know why livekit don't allow use my own agent framework ( like pydantic ai ) . There are a lot of custom configuration, like tool calling . Why not be more framework agnostic ?

This message was deleted.

Hi team, We are facing interruption handling issues with LiveKit Agents and wanted to understand the recommended architecture here. Setup: • Self-hosted LiveKit server on AWS Mumbai (

ap-south-1

) • Using LiveKit Cloud credentials (Build Plan) • Initially using

vad

interruption handling Issue: With normal

vad

mode, the agent occasionally got stuck during conversations even after configuring:

resume_false_interruption = 0.8

So we switched to

adaptive

interruption handling. However, after testing adaptive mode, we observed: 1. Genuine interruptions are sometimes missed, especially in noisy environments 2. Frequent timeout/fallback errors:

adaptive interruption inference timed out after 0.7s
fallback to VAD
status_code=408

From logs, it seems adaptive inference is happening on:

agent-gateway.livekit.cloud

So the flow becomes: Mumbai Server → LiveKit Cloud (US/EU) → Mumbai This cross-region round trip appears to exceed the strict

0.7s

timeout budget under slight latency/noise conditions. 3. Additionally, it seenms When adaptive inference times out and fallback to VAD happens, the fallback VAD does not appear to properly respect/load

min_interruption_duration

. As a result: • interruption detection becomes unstable • short/noisy audio chunks trigger inconsistent states • the agent sometimes gets stuck in an unusable state ("state 0" behavior on our side) This makes the fallback path unreliable in production. Questions: 1. Is this a known limitation/bug with adaptive → VAD fallback handling. Is there any workaround/recommended configuration for stable fallback behavior? 2. Is there any way to self-host the adaptive interruption inference/VAD model on our own infrastructure? 3. Since we already have LiveKit Cloud Build Plan credentials, is there a recommended low-latency setup for hybrid deployments (self-hosted LiveKit + cloud adaptive inference)? Currently, local

vad

mode feels more stable than adaptive mode due to these timeout/fallback issues. Would appreciate guidance on the recommended production architecture and whether this fallback behavior is a known issue.

# def llm_wrapper(metrics: LLMMetrics): # asyncio.create_task(on_llm_metrics(metrics)) # # async def on_llm_metrics(metrics: LLMMetrics): # logger.info(f"Received metrics llm: {metrics}") # # session.llm.on("metrics_collected", llm_wrapper) # # def tts_wrapper(metrics: TTSMetrics): # asyncio.create_task(on_tts_metrics(metrics)) # # async def on_tts_metrics(metrics: TTSMetrics): # logger.info(f"Received metrics tts: {metrics}") # # session.tts.on("metrics_collected", tts_wrapper) # def stt_wrapper(metrics: STTMetrics): # asyncio.create_task(on_stt_metrics(metrics)) # # async def on_stt_metrics(metrics: STTMetrics): # logger.info(f"Received metrics stt: {metrics}") I can use per plugin instance to subscribe over the metrics_collected event, as sessiion level metrics collected event is deprecated. but how will I track eou? there is no plugin instance to track eou

We're having issues running gpt-realtime-2 on LiveKit - anyone else experiencing this? For example we can't specify reasoning level, and the model seems to have weird outputs

Hi @refined-appointment-81829, @lively-winter-88668, @tall-belgium-91876 I’m trying to collect per-plugin metrics for a realtime model session. I understood: Per-plugin metrics → subscribe to the realtime model’s metrics_collected event on the plugin instance I’m currently using a llm.RealtimeModel instance through:

@property

def llm(self) -> llm.LLM | llm.RealtimeModel | None:
    return self._llm

I tried: session.llm.on(“metrics_collected”) but llm.RealtimeModel does not seem to expose

rtc.EventEmitter

methods, so .on(…) is not available. What is the correct way to subscribe to metrics_collected events for a RealtimeModel? Could someone share an implementation snippet for collecting per-plugin metrics from a realtime model session? Thanks!

What happens if the concurrency goes above our plan? I have two $50 plans, I can see a lot of times that I get charged for going above the $50 thresholds for certain things like egress, which is perfectly fine, just wanted to make sure that livekit will also handle high concurrency and not stop it and will simply just charge me for it? Also what are the overage charges for concurrency etc. I occasionally have bursts where i have 1000+ concurrent users so don't want the UX going bad.

Hey team — hitting a reliable failure on Wi-Fi network switches with livekit-agents (Python). Two failure modes from the same trigger: Stuck-in-reconnecting — connection_state_changed fires with state 2, never returns to 1. A TCP probe to the LK host stays reachable (30–100ms RTT) throughout the 30+ seconds the SDK is wedged, so it's not the network — the reconnect logic is stuck at the application layer. Ghost-room — connection_state_changed fires with state 1 (claims reconnected) but remote_participants is the stale pre-blip snapshot — no audio frames flow and participant_disconnected never fires for the participant that's actually gone. User-visible outage on a brief network blip is ~60s. We're currently working around it by monitoring connection_state_changed and force-evicting via RoomService.RemoveParticipant when the SDK doesn't recover, but would really appreciate guidance on what the SDK is supposed to do on the recovery path, or whether a fix is planned. Happy to share logs / repro steps if helpful.

Hi Team, Question: how do you make a Realtime agent speak a configured greeting verbatim at the start of a phone call? Setup: -

@livekit/agents@1.4.2

,

@livekit/agents-plugin-openai@1.4.2

- Single

voice.AgentSession({ llm: new openai.realtime.RealtimeModel({ model: 'gpt-realtime', voice: 'marin', ... }) })

— no separate STT/TTS (Realtime model is the whole pipeline) - SIP inbound calls via LiveKit - Each company has a configured first-line greeting (e.g.

"Hi, thank you for calling Chucky Cheese, this is Tony — you are on a recorded line. How can I help?"

) that we need spoken verbatim when the call connects, before any user input. What I've tried: 1.

session.say(greeting, { allowInterruptions: true })

→ throws

"trying to generate speech from text without a TTS model"

because we don't configure a separate TTS pipeline (the realtime model owns audio). 2.

session.generateReply({ instructions: greeting })

→ the model hallucinates a different greeting (e.g. says "Welcome to fun zone" for the Chucky Cheese prompt).

instructions

is a system-prompt-style directive for the turn, not text to speak. 3.

session.generateReply({ instructions: "With an upbeat tone, immediately say "${greeting}". Do not say it multiple times." })

→ mostly works, but the model occasionally leaks fragments of the wrapper meta-text (e.g. callers hear "with an upbeat attitude…" before the actual greeting). Security/UX issue for us since we're customer-facing. Is there a canonical pattern for "Realtime agent speaks this exact text first, then proceeds normally"? I've thought about: - Pre-populating the

Agent.chatCtx

with a synthetic

ChatMessage({ role: 'assistant', content: greeting })

and then calling

session.generateReply()

with no args — but I'm not sure the realtime model will actually speak a context-only item. - Mounting a separate

tts:

on the session purely so

session.say()

works — feels wasteful since the realtime model already does audio. - Adding

voice.Agent.onEnter()

override — but unclear what the right hook is for "speak this verbatim." Is there a method on

voice.AgentSession

or

realtime.RealtimeModel

I'm missing for "play this exact text through the realtime model's audio" without going through the LLM-instruction path? Pointers to a sample repo would be great.

Hi Team<

We want to create a live kit AI voice outbound calling with WebSockets

Hi Team, I am facing an issue with the LiveKit agent worker and would appreciate your guidance. Case 1 When I dispatch an agent to join a room, sometimes it works correctly, but other times only the dispatch is created while the agent worker does not respond and does not start transcribing data. In those cases, the real-time meeting continues, but the agent never starts processing, which causes me to lose the meeting data. Case 2 To resolve this issue, I implemented a Redis-based approach. What I did: • During dispatch, I check a Redis flag. • If the flag is not equal to

1

, I retry the dispatch and create a new dispatch repeatedly. • On the agent worker side, whenever the worker executes successfully or starts responding, I set the Redis flag to

1

. This implementation helped me verify whether the worker actually started and allowed me to dispatch accordingly. However, now I am facing another issue: sometimes the agent suddenly leaves the meeting even though the session is still active and was not closed. Case 3 For a more permanent solution, I contacted LiveKit AI support. They suggested replacing the Redis approach with

job_status

from LiveKit. I updated both my dispatcher and agent worker code accordingly. What I observed: • Sometimes after dispatching, I receive

job_status = 2

, which indicates the agent is running correctly. • However, most of the time I receive

job_status = 1

. When the status remains

1

, the agent does not start the job, and because of that I lose the meeting conversation data. Example logs: Room ensured: 0u4r9n Found existing dispatch: AD_8Chk8zceLc7t Creating NEW dispatch for room 0u4r9n Dispatch created: AD_X2XvFEEeMZeL Poll 5 → Status: 1 Could you please help me understand what I might be doing wrong and suggest the correct permanent solution for handling this issue? Thank you.

c est quoi la latence de mistrall tts avec le plugin en mode api ?

I want to integrate livekit with MQube. How can we get the SIP gateway address ?

Can I host the same agent for call as well as on website?

Hello, first just wanted to say livekit-agents is pretty great! Awesome work! Are there any plans on supporting us-west-2 for the livekit cloud, I only see 3 available regions in the docs https://docs.livekit.io/deploy/admin/regions/agent-deployment/?

archived the channel