Hi, i can't find prices of xai voice agent api. it is relatively new. and i can't find it on livekit or xai website. can someone tell me

Hi friends. Our ai receptionist latency is 3-4 seconds, is it tolerable? GPT says no. it is built for restaurants to take food orders.

While using Livekit Inferences, we are getting very unstable LLM TTFT which is ranging from 300ms to >5s. This is happening with over 20% of the calls right now. The Agent is deployed in LiveKit Cloud. Is there a way to mitigate this? Can we have something which is stable. The two models that I've used is gpt-4.1-mini(openai) and the other is kimi-k2 (baseten). Any help regarding this is much appreciated.

anyone is using evenlabs STT Scribe V2 Realtime? It is unstable. Sometimes, i can hear response and sometimes not.

Hi all, how to make sure we subscribe only 10-25 video/screenshare track at max in a 100-500 rooms?

can someone explain to me what this means ?

Hello everyone, My name is Rehan, and I’m from Pakistan. I recently started working with LiveKit for voice-based multi-agent development. I’m currently building my own website and have integrated the LiveKit embed to enable a voice agent directly through the site. At the moment, I’m using the LiveKit free plan. However, I’m running into an issue where my voice agent is showing the error: “Failed to join room.” I would really appreciate it if someone could guide me on where to check for the root cause of this issue. Specifically: • Should I check token generation or expiration? • Could this be related to room configuration? • Are there any logs (dashboard or server-side) that would help identify the exact problem? • Are there any free-plan limitations that might cause this behavior? Any suggestions on how to properly debug this would be greatly appreciated. Thank you in advance for your help!

👇 Please forgive the repeated reposts, but Slack is terrible about keeping messages like this in context:👇

Hi everyone, please be aware that we have a new community forum at community.livekit.io and the plan is to move support for this channel into that new forum. I understand such a move will not happen overnight, but LiveKit staff will start to prioritise the new forum when answering community support questions. Please see full details at https://livekit-users.slack.com/archives/C01KZ61HZRR/p1770299394024089, which is also the best place to ask questions and raise concerns.

Hi LiveKit team 👋 We are currently running a production application on LiveKit Cloud but we’ve lost track of the email used to create the account. Our free credits are exhausted and we need to upgrade urgently. We still have valid project credentials and the system is live. Could someone help us identify the owner account or trigger access recovery so we can upgrade the plan? We can securely share the Project URL and API Key via DM. [00:41] Thanks in advance

I need to some support to over from bottle neck issue, which is calls is not working saying connecting, Its new setup i did, and migrated livekit from VM based to K8s cluster. context: -------- "In the livekit design, If I am using haproxy loadbalancer beforw livekit node to manafer wss sigling and turn traffic via haproxy instead directly to livekit nodes, I have done all the setup and I deployed new Livekit setup in kubernetes , where Livekit is working with my existing deployed chat application call MMS service running os same cluster in different namespace, which is abstraction layer for managing the token, So architecture point of view I am using client (external) --------haproxy loadbalancer (wss or signaling sni media.example.com + WebRTC sni media-ntunr.external.com connection on 3478 (UDP)) -------> istioingressGateway + Virtualservice, I have created two services into the istio-system for service one service called X4-ingressgateway which is recieving the all TCPrequest + other livekit-ingressGateway service for livekit traffic, Like wss(signaling) on port 443 TCP is getting managed by livekit-istioingressgateway -------> livekit-gateway which two endpoint one for livekit-quartz.external.com for (WSS request) and other route livekit-nturn-quartz.external.com (turn request on port 3478/UDP) connection with policy externalTrafficPolicy: Local or externalTrafficPolicy: Cluster. currently livekit cluster is running and integrated by chat service, also I am not using public IP directly NAT with livekit nodes, Istead I am reusing the Haproxy Public IP and configure Haproxy public IP in the livekit config.yaml in RTC section in my livekit version v1.9.4 use_external_ip: true also external_ip: "Public_IP_Of _Haproxy", - Currently I have deployed the livekit into the K8s as daemonset along with redis+Sentinel cluster. --------- Architecture ------------------ Model A – LiveKit Behind L4 LoadBalancer (Correct) HAProxy passes: TCP 443 → WSS Signaling UDP 3478 → LiveKit TURN (From haproxy --- Livekit node on 3478 direct via hostNetwork) UDP media ports 40k - 50k → (From haproxy --- Livekit node on 40k - 50k LiveKit advertises HAProxy public IP BUT HAProxy must be a pure L4 passthrough please provide the haproxy changes which should work with UDP, Note I am using Haproxy enteprise version which is supporting UDP traffic now.] current haproxy config ------------------------ global maxconn 10000 log 127.0.0.1 local0 log 127.0.0.1 local1 notice chroot /var/empty stats socket /run/hapee-3.0/hapee-lb.sock mode 660 level admin expose-fd listeners stats timeout 10m module-path /opt/hapee-3.0/modules module-load hapee-lb-udp.so daemon nbthread 1 #cpu-map 1 0 #cpu-map 2 1 # ======================= # DEFAULTS # ======================= defaults log global option tcplog option dontlognull option redispatch option logasap timeout connect 5000 timeout client 50000 timeout server 50000 # ======================= # HTTPS / WSS / TURN-TLS FRONTEND # ======================= frontend https bind 0.0.0.0:443 mode tcp tcp-request inspect-delay 5s tcp-request content accept if { req_ssl_hello_type 1 } # ---- Matrix / App traffic ---- acl acl_synapse_matrix req_ssl_sni -i quartz1.service421.com acl acl_ntfy_matrix req_ssl_sni -i quartz1-ntfy.service421.com acl acl_sygnal_matrix req_ssl_sni -i quartz1-sygnal.service421.com use_backend bk_istio_ingress_gw if acl_synapse_matrix use_backend bk_istio_ingress_gw if acl_ntfy_matrix use_backend bk_istio_ingress_gw if acl_sygnal_matrix # ---- LiveKit ---- acl acl_livekit_wss req_ssl_sni -i livekit-quartz.service421.com acl acl_livekit_turn req_ssl_sni -i livekit-nturn-quartz.service421.com use_backend bk_livekit_wss if acl_livekit_wss use_backend bk_livekit_tls if acl_livekit_turn # ======================= # BACKENDS # ======================= # ---- General Istio Gateway (apps) ---- backend bk_istio_ingress_gw mode tcp server aw01-quartz aw01-quartz:32555 check server aw02-quartz aw02-quartz:32555 check server aw03-quartz aw03-quartz:32555 check # ---- LiveKit WSS (443 → 30880) ---- backend bk_livekit_wss mode tcp balance leastconn option tcp-check server lk01 lk01-quartz:30880 check server lk02 lk02-quartz:30880 check server lk03 lk03-quartz:30880 check # ---- LiveKit TURN-TLS (5349 → 32550) ---- backend bk_livekit_tls mode tcp balance leastconn option tcp-check server lk01 lk01-quartz:32550 check server lk02 lk02-quartz:32550 check server lk03 lk03-quartz:32550 check # ---- Traefik PoC ---- backend bk_traefik_poc mode tcp server k3sm-quartz k3sm-quartz:30499 check # ===================================================================== # FRONTENDS (TCP) # ===================================================================== # TURN TLS frontend turn_tls_tcp_frontend bind *:5349 mode tcp option tcplog default_backend livekit_turn_tls_tcp # TURN Plain (TCP) frontend turn_tcp_frontend bind *:3478 mode tcp option tcplog default_backend livekit_turn_tcp # ===================================================================== # BACKENDS (TCP) # ===================================================================== backend livekit_turn_tls_tcp mode tcp balance source server lk01-k8s lk01-quartz:3478 check server lk02-k8s lk02-quartz:3478 check server lk03-k8s lk03-quartz:3478 check backend livekit_turn_tcp mode tcp balance source server lk01-k8s lk01-quartz:3478 check server lk02-k8s lk02-quartz:3478 check server lk03-k8s lk03-quartz:3478 check # ===================================================================== # UDP LOAD BALANCING (HAPEE UDP-LB) # ===================================================================== # TURN UDP udp-lb udp_livekit_turn dgram-bind *:3478 balance source server lk01-k8s lk01-quartz:3478 check server lk02-k8s lk02-quartz:3478 check server lk03-k8s lk03-quartz:3478 check # RTC Primary Media Port udp-lb rtc_udp_7882 dgram-bind *:7882 balance source server lk01-k8s lk01-quartz:3478 check server lk02-k8s lk02-quartz:3478 check server lk03-k8s lk03-quartz:3478 check # WebRTC RTP Media Range (40000-50000) # 1:1 port mapping is critical here #udp-lb udp_livekit_rtp # dgram-bind *:40000-50000 # balance source # server lk01 lk01-quartz check

👇 Please forgive the repeated reposts, but Slack is terrible about keeping messages like this in context:👇

Hi everyone, please be aware that we have a new community forum at community.livekit.io and the plan is to move support for this channel into that new forum. I understand such a move will not happen overnight, but LiveKit staff will start to prioritise the new forum when answering community support questions. Please see full details at https://livekit-users.slack.com/archives/C01KZ61HZRR/p1770299394024089, which is also the best place to ask questions and raise concerns.

Hello everyone, Title: High Latency in Unity + LiveKit (Mumbai) + Gemini Native Audio Streaming Hi Team, We are building a real-time AI voice assistant using the following architecture: Unity (Mic Input) ↓ LiveKit Cloud (Mumbai region) ↓ Python Agent (hosted on local machine) ↓ (streaming audio) Gemini model: gemini-2.5-flash-native-audio-preview-12-2025 ↓ (streaming audio response) Python Agent ↓ (publish audio immediately) LiveKit ↓ Unity playback Issue: We are experiencing ~3–4 seconds delay before AI response audio starts playing in Unity. Observations: - Direct Gemini API testing shows low latency. - Latency increases only when routing through Python agent + Gemini. - Audio is streamed (not waiting for full transcript). - Audio is published back immediately upon receiving chunks. Questions: 1. Are there recommended audio frame sizes for low-latency streaming with Gemini native audio? 2. Are there specific LiveKit or WebRTC buffering parameters recommended for Unity? 3. Could local agent hosting cause additional latency compared to AWS? 4. Is there a best-practice architecture for achieving sub-1.5s response start time? Our target is response audio starting within ~1–1.5 seconds. Any guidance would be greatly appreciated. Thanks.

Hi! I can help with this. From your setup, the delay is most likely coming from buffering and serialization inside the Python agent layer rather than Gemini itself, especially since direct API tests are fast. Running the agent locally can also add extra round-trip and processing overhead compared to deploying it in the same region as LiveKit. With proper tuning of frame sizes, streaming chunk relay, and LiveKit/Unity jitter buffer settings, it’s usually possible to bring response start time closer to the 1–1.5s target. I’d be happy to help profile the pipeline end-to-end and optimize each stage for lower latency. @green-area-83933

Darryn Campbell [10:35 AM] alert This channel is being sunset on 27th February 2026: alert

Hi everyone, please be aware that we have a new community forum at community.livekit.io and the plan is to move support for this channel into that new forum. Please see full details here. The Slack community will remain, but we will do housekeeping of the technical channels after that date.

Elevenlabs announced today all their default voices expire in December and they have a new list of non-expiring voices. What is the ETA for making the new voices available inside inference? I tried already and it says the voice is not available

Hi, I want to ask about best practices for TTS in LiveKit. We currently use OpenAI whisper-1 (realtime) as the STT model in our LiveKit agent to transcribe Japanese utterances, but we sometimes experience delays. Because of that, we’re planning to replace OpenAI whisper-1 with another model. What would be the best choice? Does anyone have experience with this?

Hi every one I am running LLM in open source which LLM do you suggest with function tools

we're working on it. In the meantime, do you need me to add any voices? please pass me the voice ids.

Hi, I am getting this message in all projects: Your project has exceeded its free tier limit. Please upgrade now. I also received an email regarding a specific project. If I log into the project billing dashboard or in plan quotas, everything looks OK. Can you please advise?

Subject: ETA on PR #5238? (Gemini 3.1 Flash Live generate_reply 1007 error) > > Hey LiveKit team! 👋 We are super excited to upgrade our language learning app to the new gemini-3.1-flash-live-preview model, but we are currently blocked by a WebSocket payload incompatibility. > > Right now, when our agent joins the room and tries to speak its initial greeting using session.generate_reply(), it triggers an immediate fatal WebSocket error: > 1007 (invalid frame payload data) Request contains an invalid argument > > I saw that PR #5238 (fix(google realtime): support gemini-3.1-flash-live-preview A2A model) perfectly addresses this by dropping send_client_content (the . hack) and routing text through send_realtime_input(text=...) to satisfy the new A2A model restrictions. > > However, because the currently published PyPI packages (livekit-plugins-google==1.4.5 / livekit-agents==1.5.1) still send the old payload formats, the agent just stares silently and the connection drops. We had to roll our production app back to Gemini 2.5 Flash Native. > > Question: Is there an ETA on when PR #5238 will be merged and published to PyPI as a new release? We'd love to test out the 3.1 improvements as soon as it drops! Thanks for all the hard work on the Google integration. 🚀

I'm using Sarvam for STT and TTS and I got an error that says " 204850.606 ERROR… livekit.agents error decoding WAV audio ", seems like Sarvam TTS is giving outputs in mp3. How to fix this

Hi everyone, I'm getting an average ttft of ~1,12s on gpt-4.1 calls using LiveKit Inference, but the benchmarks claim ~670ms. Thank you in advance for your help!

Hello I just have a question regarding live-streaming using this platform. How many hours do you get with shipped total?

Hi all, are there plans to make

openai/gpt-5.5-chat-latest

available via

inference.LLM

?

openai/gpt-5.3-chat-latest

maybe retires in June?

Hi everyone, I got a mail from LiveKit saying our project used 80% of the free inference credits for this month. Can someone explain what this limit actually is and which service is using it? Also, what will happen if the limit gets exceeded? We are using LiveKit for voice assessment features in our project. Just wanted to understand the warning mail better. Thanks!

Hi all! I got a message this morning that I was approaching 80% of my limit. I my voice-enabled agent instance actually stopped generating speech last night well before I got this message and have been unable to get it to respond or state it's greeting message. Are the two related or is there a problem of some kind happening right now?

How do we share Agent Builder editable access across team members within the same project?

Hi all can someone please help me with this query that I raised in the community a couple of days ago related to livekit inference. It's very critical for us and is a blocker for scaling further. https://community.livekit.io/t/data-residency-for-deepgram-stt-when-using-livekit-inference-in-india-region/1203

archived the channel