hi
# ask-aic
cuddly-vr-85751
02/15/2026, 3:58 PMhi
t
tall-airline-16756
02/15/2026, 3:58 PMThanks for reaching out, Arun. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.
c
cuddly-vr-85751
02/15/2026, 3:58 PMfrom datetime import datetime
import json
import time
from dotenv import load_dotenv
from livekit import api
from livekit.agents import Agent, JobContext, JobProcess, WorkerOptions, cli, function_tool
from livekit.agents.voice import AgentSession
from livekit.plugins import silero
from agents.functions.function_builder import function_builder
from agents.utils import get_llm_from_config, get_stt_from_config, get_tts_from_config, inactivity_handler
from agents.utils import WELCOME_MESSAGE, OUTBOUND_MESSAGE, END_CALL_PROMPT
from agents.utils import get_all_metadata, get_all_instructions, substitute_variables, get_interruption_params
from agents.services.call_recording import CallRecordingService
from agents.api import end_call
from agents.rag import get_embed_model
load_dotenv()
def prewarm(proc: JobProcess):
proc.userdata["vad"] = silero.VAD.load()
proc.userdata["embedding_model"] = get_embed_model()
class VoiceAgent(Agent):
def __init__(self, ctx, meta_data, variables=None, call_metadata=None):
self.ctx = ctx
self.rag = meta_data['rag']
super().__init__(
instructions=get_all_instructions(meta_data['agent'], variables),
stt=get_stt_from_config(meta_data['stt']),
llm=get_llm_from_config(meta_data['llm']),
tts=get_tts_from_config(meta_data['tts']),
tools=function_builder(meta_data['functions'], call_metadata)
)
async def on_user_turn_completed(self, turn_ctx, new_message) -> None:
if not self.rag:
return
query = new_message.text_content
if not query or len(query) < 3:
return
rag_content = await self.rag.query(query)
if rag_content:
turn_ctx.add_message(
role="assistant",
content=f"[Knowledge Base Context]: {rag_content}"
)
@function_tool(description=f"{END_CALL_PROMPT}")
async def end_call_on_user_request(self) -> str:
self.ctx.shutdown()
return json.dumps({"status": "session shutdown initiated"}, indent=2)
async def entrypoint(ctx: JobContext):
meta_data = await get_all_metadata(ctx)
dial_info = meta_data.get('dial_info', {})
to_number = dial_info.get('to_number')
from_number = dial_info.get('from_number')
sip_trunk_id = dial_info.get('sip_trunk_id')
variables = dial_info.get('variables', {})
is_outbound = bool(to_number)
if is_outbound:
if not sip_trunk_id:
ctx.shutdown()
return
try:
await ctx.api.sip.create_sip_participant(
api.CreateSIPParticipantRequest(
room_name=ctx.room.name,
sip_trunk_id=sip_trunk_id,
sip_call_to=to_number,
participant_identity=to_number,
participant_name=f"Call to {to_number}",
krisp_enabled=True,
wait_until_answered=True,
)
)
except api.TwirpError as e:
ctx.shutdown()
return
is_telephony = True
channel = "PHONE"
else:
await ctx.connect()
is_telephony = False
for participant in ctx.room.remote_participants.values():
if hasattr(participant, 'attributes') and participant.attributes:
attrs = participant.attributes
sip_phone = attrs.get('sip.phoneNumber')
if sip_phone:
from_number = sip_phone
to_number = attrs.get('sip.trunkPhoneNumber')
is_telephony = True
break
channel = "PHONE" if is_telephony else "WEB"
call_metadata = {
"from_number": from_number if is_telephony else None,
"to_number": to_number if is_telephony else None,
"call_type": channel,
"direction": "OUTBOUND" if is_outbound else "INBOUND" if is_telephony else None,
}
agent = VoiceAgent(ctx, meta_data, variables, call_metadata)
agent_data = meta_data['agent']
interruption_sensitivity = agent_data.get('interruption_sensitivity', 0.5)
min_interruption_duration, false_interruption_timeout = get_interruption_params(interruption_sensitivity)
session = AgentSession(
vad=ctx.proc.userdata.get("vad"),
preemptive_generation=False,
min_interruption_duration=min_interruption_duration,
false_interruption_timeout=false_interruption_timeout,
)
inactivity_handler(session, ctx)
start_time = datetime.now()
recorder = CallRecordingService()
await recorder.start_recording(
room_name=ctx.room.name,
filename=f"{ctx.room.name}_{int(time.time())}.ogg",
)
await session.start(agent=agent, room=ctx.room)
if is_telephony and is_outbound:
greeting = agent_data.get('greeting_message', OUTBOUND_MESSAGE)
else:
greeting = agent_data.get('greeting_message', WELCOME_MESSAGE)
greeting = substitute_variables(greeting, variables)
await session.say(greeting)
async def on_session_end():
recording_url = await recorder.stop_and_wait()
end_call_data = {
"session_id": ctx.job.id,
"started_at": start_time.isoformat(),
"ended_at": datetime.now().isoformat(),
"transcript": json.dumps(session.history.to_dict() if hasattr(session, "history") else {"items": []}, indent=2),
"agent_id": agent_data.get('id'),
"client_id": agent_data.get('client_id'),
"recording_url": recording_url,
"direction": "OUTBOUND" if is_outbound else "INBOUND" if is_telephony else None,
"channel": channel,
"from_number": from_number if is_telephony else None,
"to_number": to_number if is_telephony else None,
}
try:
await end_call(end_call_data)
except Exception as e:
print(f"Error logging call: {e}")
try:
ctx.delete_room(ctx.room.name)
except Exception as e:
print(f"Error deleting room: {e}")
ctx.add_shutdown_callback(on_session_end)
def run():
cli.run_app(
WorkerOptions(
entrypoint_fnc=entrypoint,
prewarm_fnc=prewarm,
agent_name="voice-agent",
port=8082,
num_idle_processes=2,
)
)
if name == "__main__":
run() thsi si deployed on aws vm but i fee leiktt high lentency
t
tall-airline-16756
02/15/2026, 3:58 PMHi — great to see you here! How can I help today?
If you’re not sure where to start, here are a few things I can assist with:
• Troubleshooting a LiveKit room or connection issue
• SDK integration questions (Web, iOS, Android)
• Token/credentials or permissions help
• Performance or quality problems (audio/video)
• Billing, plans, or account questions
Tell me which area you’re working on and any details (project/room IDs, error messages, SDK & version) and I’ll jump in.
tall-airline-16756
02/15/2026, 3:59 PMHi — thanks for sharing your code and letting me know you’re seeing high latency. I’m here to help. I’ll walk through the most likely causes based on your deployment (AWS VM) and the code you posted, then ask a couple of quick questions so I can give more precise recommendations.
What I’m responding to: your voice-agent code deployed on an AWS VM and the note “i fee leiktt high lentency” (high latency).
Most likely causes and what you can do now
• EC2 instance type (highest likelihood)
◦ If you’re on a burstable instance (t3/t4 family like t3.micro, t3.small), CPU credit exhaustion causes drastic throttling that looks like “high latency” even with low reported CPU. Check CloudWatch for
CPUCreditBalancepreemptive_generation=Falsepreemptive_generation=True Copy code
session = AgentSession(
vad=ctx.proc.userdata.get("vad"),
preemptive_generation=True, # reduces perceived latency
min_interruption_duration=min_interruption_duration,
false_interruption_timeout=false_interruption_timeout,
)krisp_enabled=TrueCPUCreditBalanceCPUCreditBalancec
cuddly-vr-85751
02/15/2026, 3:59 PMt3 meduis
t
tall-airline-16756
02/15/2026, 4:00 PMThanks — that’s a very helpful detail. A t3.medium is a burstable instance and is likely the main cause of the high latency you’re seeing. Here’s a short, prioritized plan you can follow right now to confirm and fix the issue.
What I’m responding to: you’re running the voice agent on an AWS t3.medium and experiencing high latency.
Why t3.medium is likely the problem
• t3 instances use CPU credits. When credits run out the instance is throttled to a small fraction of a CPU, which causes long pauses even though reported CPU can look low.
• This throttling commonly shows up as slow VAD/STT/LLM/TTS response times and long turn latency.
Immediate checks (do these first)
• Check CloudWatch for the
CPUCreditBalancekrisp_enabled=Falsepreemptive_generation=TrueAgentSessionCPUCreditBalancec
cuddly-vr-85751
02/15/2026, 4:09 PMeven in lcola alos slow ahat sod i do to amke fast ets
t
tall-airline-16756
02/15/2026, 4:09 PMThanks — I hear your frustration. You said it’s slow even locally. Let’s narrow down where the delay is and fix it step‑by‑step. I’ll give a short diagnostic checklist you can run now, plus quick fixes to try immediately.
What I’m responding to: your agent is high-latency both on t3.medium (AWS) and when running locally.
Quick diagnostic checks (run these first)
1. Time each pipeline stage with simple logs:
◦ Log timestamps for: VAD end detection, STT request start/end, LLM request start/end (TTFT), TTS request start/end (TTFB), and audio playback start.
◦ This tells you whether latency is in VAD, STT, LLM, TTS, or network.
1. Test a minimal local loop:
◦ Replace real STT/LLM/TTS with a stub that immediately returns a small response. If this is fast, the external providers are the bottleneck.
1. Network RTT test:
◦ From your machine and from the VM, ping/traceroute the STT/LLM/TTS endpoints and measure round-trip time.
1. Resource check (local & VM):
◦ CPU, RAM, disk I/O while running a call. On AWS, check CloudWatch
CPUCreditBalancepreemptive_generation=TrueAgentSessionpreemptive_generation=False2 Views