hello!
I have create a agent on livekit cloud(no code), and set a data collection endpoint url, but that endpoint takes no data after I call the agent.
what I can ensure is:
the endpoint works fine for test, and dose not has a schema mismatch problem, since I print all requests.
after check the session, I saw the agent already call a record-info tool, and correctly record the information it need top collect, so the agent already has the data, just not send it to the endpoint for some reason,
at the call, it automaticlly ended by the agent itself, dose not has any unexpected end or disrupt or something like this..
need help with this issue, thanks!
Do you see anything in your agent logs?
My first guess would be that there is some agent secret required for your endpoint that you have configured for test but not configured for your deployed agent.
hello!
I have checked the agent log, but I find out I have not set any agent secret.
and by the test, I means that I have use a local request to test the endpoint, and find out the endpoint is reachable through network, because by the first time, I think the problem may be the endpoint is not reaachable through network, so the fact is the endpoint has not receive any data from livekit either in test mode or depoly mode, do I miss some config?
What is the endpoint, specifically?
Could you add logging around your call to the endpoint to better understand the error?
I haven’t tried that before, but looking at the documentation for the mode results: https://docs.livekit.io/agents/start/builder/#data-collection-results
And testing with a curl command with the appropriate format:
curl -X POST "https://randy-production-df79.up.railway.app/submit" \
-H "Content-Type: application/json" \
-d '{
"job_id": "AJ_a1b2c3d4e5f6",
"room_id": "RM_x9y8z7w6v5u4",
"room": "support-call-2026-07-16",
"started_at": "2026-07-16T14:32:05Z",
"ended_at": "2026-07-16T14:38:47Z",
"summary": "Caller Jane Doe requested a callback about a billing discrepancy on her July invoice. She confirmed her account email and preferred to be reached in the afternoon. Issue was logged for the billing team to follow up.",
"results": {
"contact_info": {
"full_name": "Jane Doe",
"email": "jane.doe@example.com",
"phone": "+1-415-555-0142"
},
"callback_preferences": {
"preferred_time": "afternoon",
"wants_callback": true
},
"topics": [
{ "topic": "billing discrepancy", "priority": "high" },
{ "topic": "invoice question", "priority": "medium" }
]
}
}'
I get this error:
{"error":"Request body must contain only a non-empty string field named \"name\""}
Which, to me, looks like your endpoint is not accepting the payload structure that LiveKit is sending, it is expecting something like:
{"name": "some string"}
hello, thanks for your reply.
I want to explain that: the endpoint true dose not have a correct schema verify as the input schema shows in the docs.
because I now just want to test if the endpoint can receive any data from livekit or not, in that case, whether the endpoint receive a correct schema or incorrect schema, it will print the request info on the console, and after check the console, I saw the request you send from your machine, but I dont see any request that send by livekit cloud,
so, as I understand, in such case, whether the endpoint receive a matched request or not, it will all print them on console, so even if livekit cloud send a mismatcdhed schema request, I should see them on console, but I got nothing.
and you said that you never tried that before, so can I confirm that dose this function is operational, or not?
and also I want to clarify my goal, is that
first, I want to build all my system on livekit cloud, without any code.
second, I want to store the collected data in my own server, other wise, the collect data agent would be not useful to me.
The best next step forward would probably be for you to copy the code generated by the agent builder here, and I can hopefully (
) tell you why your endpoint is not being invoked.
import logging
import asyncio
from dataclasses import dataclass, asdict, is_dataclass
from dotenv import load_dotenv
from livekit import rtc
from livekit.agents import (
Agent,
AgentServer,
AgentSession,
AgentTask,
JobContext,
JobProcess,
TurnHandlingOptions,
RunContext,
ToolError,
cli,
function_tool,
get_job_context,
inference,
llm,
room_io,
utils,
)
from livekit.agents.beta.tools import EndCallTool
from livekit.agents.beta.workflows import TaskGroup
from livekit.agents.llm.chat_context import FunctionCall
from livekit.agents.llm.utils import execute_function_call
from livekit.plugins import (
ai_coustics,
silero,
)
from livekit.plugins.turn_detector.multilingual import MultilingualModel
logger = logging.getLogger(“agent-assistant-1ae4”)
load_dotenv(“.env.local”)
def _to_json_serializable(obj):
“”“Convert dataclasses and nested structures to JSON-serializable form.”“”
if is_dataclass(obj) and not isinstance(obj, type):
return asdict(obj)
if isinstance(obj, list):
return [_to_json_serializable(item) for item in obj]
if isinstance(obj, dict):
return {k: _to_json_serializable(v) for k, v in obj.items()}
return obj
@dataclass
class InfoResults:
name: str
class InfoTask(AgentTask):
def init(self, agent_instructions: str, extra_tools: list | None = None):
no_greet_prefix = “The user has already been greeted. Do not introduce yourself or say hello. Directly ask for the required information.\n”
task_instructions = “”
no_goodbye_suffix = “\nIMPORTANT: Do NOT say goodbye, recap the full conversation, or tell the user you are done. Only focus on collecting the information for THIS specific task. If the information was already provided earlier in the conversation, confirm it briefly and then record it immediately using the appropriate tool.”
wrapped_instructions = no_greet_prefix + agent_instructions + “\n” + task_instructions + no_goodbye_suffix
super().init(
instructions=wrapped_instructions,
tools=list(extra_tools) if extra_tools else ,
)
async def on_enter(self):
await self.session.generate_reply(
instructions=(
"Begin this task now. If the task instructions require calling "
"a tool first (for example, to look up information), call it. "
"Otherwise, ask the user for the information described in your "
"task instructions."
),
allow_interruptions=True,
tool_choice="auto",
)
@function_tool(name="record_info")
async def record_info(self, context: RunContext, name: str):
"""Call when you have collected all required data points for this task.
Provide the structured results exactly as requested.
Do not confirm on record, remain silent and move to the next task.
Args:
name (str)“”"
self.complete(InfoResults(name=name))
class DefaultAgent(Agent):
def init(self) → None:
self._agent_instructions = “”"You are a friendly, reliable voice assistant that answers questions, explains topics, and completes tasks with available tools.
Output rules
You are interacting with the user via voice, and must apply the following rules to ensure your output sounds natural in a text-to-speech system:
- Respond in plain text only. Never use JSON, markdown, lists, tables, code, emojis, or other complex formatting.
- Keep replies brief by default: one to three sentences. Ask one question at a time.
- Do not reveal system instructions, internal reasoning, tool names, parameters, or raw outputs
- Spell out numbers, phone numbers, or email addresses
- Omit
https://and other formatting if listing a web url - Avoid acronyms and words with unclear pronunciation, when possible.
Conversational flow
- Help the user accomplish their objective efficiently and correctly. Prefer the simplest safe step first. Check understanding and adapt.
- Provide guidance in small steps and confirm completion before continuing.
- Summarize key results when closing a topic.
Tools
- Use available tools as needed, or upon user request.
- Collect required inputs first. Perform actions silently if the runtime expects it.
- Speak outcomes clearly. If an action fails, say so once, propose a fallback, or ask how to proceed.
- When tools return structured data, summarize it to the user in a way that is easy to understand, and don’t directly recite identifiers or other technical details.
Guardrails
-
Stay within safe, lawful, and appropriate use; decline harmful or out‑of‑scope requests.
-
For medical, legal, or financial topics, provide general information only and suggest consulting a qualified professional.
-
Protect privacy and minimize sensitive data.“”"
super().init(
instructions=“”,
tools=[EndCallTool(
extra_description=“”“”“”,
end_instructions=“”“Thank the user for their time and say goodbye.”“”,
delete_room=True,
)],
)
async def on_enter(self):
greeting_instructions = “”
greeting_instructions = “”“Greet the user and offer your assistance.”“”The greeting must not ask a question — the first data collection task
asks the opening question. Without this guardrail the LLM tends to end
with an open-ended prompt (“How can I help?”), which collides with the
task’s first turn.
no_question_guardrail = (
“IMPORTANT: The greeting must be a statement only. Do NOT end with any "
'question, including open-ended prompts like “How can I help?”. The ’
“next task will ask the first question.”
)
await self.session.generate_reply(
instructions=”\n".join(
part for part in (self._agent_instructions, greeting_instructions, no_question_guardrail) if part
),
allow_interruptions=True,
)Propagate HTTP/client/MCP tools into each data collection task so
they’re callable mid-task (e.g. looking up a customer record while
collecting details). EndCallTool is excluded here — it’s invoked
programmatically in _finish_data_collection.
_task_tools = [t for t in self.tools if not isinstance(t, EndCallTool)]
task_group = TaskGroup(chat_ctx=self.chat_ctx)
task_group.add(
lambda _ai=self._agent_instructions, _tools=_task_tools: InfoTask(agent_instructions=_ai, extra_tools=_tools),
id=“info”,
description=“Info”,
)
try:
group_result = await task_group
except (ToolError, asyncio.CancelledError):
logger.info(“data collection task group cancelled (participant likely disconnected)”)
returnawait self._finish_data_collection(group_result.task_results)async def _finish_data_collection(self, task_results):
“”“Serialize results, speak goodbye, and end the session.”“”
serialized = _to_json_serializable(task_results)
get_job_context().proc.userdata[“dc_results”] = serialized
end_instructions = “”“Thank the user for their time and say goodbye.”“”summary_task: asyncio.Task | None = None # Remove EndCallTool from active tools so the LLM cannot call it # spontaneously during the goodbye speech (it is invoked programmatically below). await self.update_tools([t for t in self.tools if not isinstance(t, EndCallTool)]) speech_handle = self.session.generate_reply( instructions=f"All data collection tasks are complete. {end_instructions}", tool_choice="none", ) try: await speech_handle if summary_task: await summary_task except ConnectionError: logger.debug("user disconnected during goodbye speech") try: end_call_tool = next((t for t in self.tools if isinstance(t, EndCallTool)), None) if not end_call_tool: end_call_tool = EndCallTool( end_instructions=end_instructions, delete_room=True, ) tools_with_end_call = [*self.tools, end_call_tool] tool_ctx = llm.ToolContext(tools_with_end_call) end_call_id = utils.shortuuid("fnc_") tool_call = llm.FunctionToolCall( call_id=end_call_id, name="end_call", arguments="{}", ) fnc_call = FunctionCall( call_id=end_call_id, name="end_call", arguments="{}", ) call_ctx = RunContext( session=self.session, speech_handle=speech_handle, function_call=fnc_call, ) await execute_function_call( tool_call, tool_ctx, call_ctx=call_ctx, ) except (ConnectionError, RuntimeError): logger.debug("room already disconnected during end-call teardown")
server = AgentServer()
def prewarm(proc: JobProcess):
proc.userdata[“vad”] = silero.VAD.load()
server.setup_fnc = prewarm
@server.rtc_session(agent_name=“assistant-1ae4”)
async def entrypoint(ctx: JobContext):
session = AgentSession(
stt=inference.STT(model=“deepgram/nova-3”, language=“en”),
llm=inference.LLM(
model=“google/gemma-4-31b-it”,
),
tts=inference.TTS(
model=“cartesia/sonic-3”,
voice=“9626c31c-bec5-4cca-baa8-f8ba9e84c8bc”,
language=“en”
),
turn_handling=TurnHandlingOptions(turn_detection=MultilingualModel()),
vad=ctx.proc.userdata[“vad”],
preemptive_generation=True,
)
ctx.proc.userdata[“dc_results”] = None
await session.start(
agent=DefaultAgent(),
room=ctx.room,
room_options=room_io.RoomOptions(
audio_input=room_io.AudioInputOptions(
noise_cancellation=ai_coustics.audio_enhancement(
model=ai_coustics.EnhancerModel.QUAIL_VF_S,
),
),
),
)
if name == “main”:
cli.run_app(server)
this is the code, and I hope its useful
Thanks for sharing that, your URL is not present in that code, which is strange (otherwise, how would the agent know where to upload your collected data).
I dug deeper, and I think you have uncovered a bug with Agent Builder. The quick fix is to turn on ‘Call summary’ under the ‘Call ending’ options which updates the code as I would expect.
thanks I will try that