how can i recover from this error that I m catching in sessi LiveKit Community #ask-ai

how can i recover from this error that I'm catchin...

aloof-scooter-47164

12/22/2025, 2:06 AM

how can i recover from this error that I'm catching in @session.on(“error”), it's making that no no response gets generated

Copy code

19:58:43.616 DEBUG  livekit.…ns.openai <<< {'type': 'error', 'event_id': 'event_CS', 'error': {'type': 'invalid_request_error',      
                                           'code': 'conversation_already_has_active_response', 'message': 'Conversation already has an active response in   
                                           progress: resp_. Wait until the response is finished before creating a new one.', 'param':  
                                           None, 'event_id': None}}                                                                                         
                                           
                 ERROR  livekit.…ns.openai OpenAI Realtime API returned an error  
                                           {"error": "RealtimeError(message='Conversation already has an active response in progress: 
resp_CmSrSgYHYc4spjMtDTWnf. Wait until the response is finished before creating a new one.', type='invalid_request_error', 
code='conversation_already_has_active_response', event_id=None, param=None)",
    19:58:43.617 DEBUG  events             Recoverable error: OpenAI Realtime API returned an error (body=RealtimeError(message='Conversation already has   
                                           an active response in progress: resp_. Wait until the response is finished before creating  
                                           a new one.', type='invalid_request_error', code='conversation_already_has_active_response', event_id=None,       
                                           param=None), retryable=True)

tall-airline-16756

12/22/2025, 2:06 AM

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

tall-airline-16756

12/22/2025, 2:06 AM

Thanks — I’m responding to your error about

conversation_already_has_active_response

that’s preventing new replies. Why this happens (brief) • The model is already producing a response (audio/text). The Realtime API prevents starting another response until the active one finishes. What you can do (practical steps) 1. Wait for the current response to finish before calling

generate_reply()

again. ◦ Use the

SpeechHandle

returned by

generate_reply()

say()

and

await

it so you only start the next reply after it completes. 1. Guard calls so you don’t start concurrent replies. ◦ Check agent state before generating:

Copy code

if session.agent_state != voice.AgentState.IDLE:
         return  # or queue the request
     await session.generate_reply(...)

• Or serialize with a mutex/lock:

Copy code

from asyncio import Lock
     speech_lock = Lock()
     async def safe_generate(instructions):
         async with speech_lock:
             handle = session.generate_reply(instructions=instructions)
             await handle

1. Avoid duplicate triggers / race conditions ◦ Deduplicate events that call

generate_reply()

(RPC handlers, message events, UI clicks). ◦ Add short debounce or rate-limit so repeated events don’t create multiple calls. 1. Handle interruptions and state events ◦ If a user interrupts, wait until interruption finishes then start new response. ◦ Listen for

agent_speech_finished

agent_state_changed

and only generate when state is IDLE. 1. Logging & debugging ◦ Log each

generate_reply()

call with timestamps and unique IDs to find where duplicates originate. If you want, share: • the snippet where you call

generate_reply()

(or your

@session.on("error")

handler), • whether you use events like

agent_state_changed

agent_speech_finished

aloof-scooter-47164

12/22/2025, 2:10 AM

but I'm not generaron any response manually, is part of the LiveKit agent code

tall-airline-16756

12/22/2025, 2:10 AM

Thanks — I hear that the agent code (not you) is triggering responses. To help pinpoint and fix it I need a bit more info. Could you share: • The LiveKit agent version you’re running. • The code paths or event handlers where the agent triggers responses (snippets showing where

generate_reply()

say()

is called, or your

@session.on(...)

handlers). • Any customizations to the agent (middleware, RPC handlers, auto-responses, or queued prompts). • Logs around when the first response starts and when the error occurs (timestamps for both). In the meantime, quick checks you can do now: • Verify the agent listens for

agent_state_changed

agent_speech_finished

and only issues new replies when state is IDLE. • Search the agent code for any places that call `generate_reply()`/`say()` automatically and add a guard like a lock or a check for

session.agent_state == voice.AgentState.IDLE

before calling. • Add logging around every internal generate call (unique IDs + timestamps) to see overlapping calls. Send the code/log snippets and agent version and I’ll point out exactly where to add guards or fixes.

2 Views

Open in Slack

Previous Next