Hi Vapi team, We've got a weird issue where our bot's just start repeating themselves within a node, example call ID is: c3312a7c-64c3-458c-9904-3c51d06bec84 First reported by one of our managers testing one of our other Workflows.

Looking into it

Thanks mate, appreciate it

I have been having this issue as well. Sometimes it repeats the whole sentence/idea multiple times in a row, but more often it starts a sentence and gets a few words in and then starts it again

yes me too

@Daco, @dakota, @Aibaze We’ve just deployed a fix that should resolve the issue. Could you please try again and let me know how it goes? - If the problem persists, please send me the new call ID and I’ll jump on it right away. - If everything looks good, just confirm and I’ll mark this as resolved. Thanks for your patience!

Sorry mate, i'm still getting random repeats. In addition i'm absolutely struggling with making any successful API calls withint he Workflow bots - is this production ready? Call ID: 9fc79c16-1533-4d9f-a6c4-6197f7b3da3b Repeats herself, simple API request fails but i can't see any API output in logs to see what/where it went wrong? Any assistance, greatly appreciated

Can you use gpt-4o model?

Pretty sure i am/was using 4o

Hey, Would you be available to schedule a brief call at your earliest convenience so that we may discuss this matter in detail? Please use the following link to book a suitable time:

Hey Sahil, i've provided information for troubleshooting, is the meeting required to get more information? Happy to meet if it's required

I am also still getting the repeating issue, just not as frequently. Also using gpt-4o model

Hey, can you send me a screen recording video that walks through the issue you are facing? It would be easier for me to understand your issue and help you out with it.

I am also experiencing the same issue and need this resolved urgently. We are using the gpt-4o model. Agent starts repeating itself randomly throughout conversation: Call ID: aff34aec-6d72-4d54-830f-7ddf5cf09b38 Timestamp: 1:25

Hey Sahil, as per James' feedback, there's not really anything a screen recording will show that the logs and transcript don't

Hey @Damir, Does the repeating issue start ocurring after an API call function is made by any chance? My team has noticed a pattern that the issue seems to start after this tool is used.

Hey @James , pretty sure the most common are after an API call, but have seen it in the absence of one too. We are strictly using 11labs for Australian accent.. so that would be disappointing if it is an issue with them

I 100% agree, we are using 11labs for the same reason too - I can definitely confirm it seems to be an issue between VAPI and 11labs currently - if you switch to VAPI as your voice provider temporarily does this fix the repeating?

I'll suss it out when I get some time today

Easy fingers crossed it works for you

Thanks mate, just on that, have you had any success changing the model 11labs uses? e.g. Turbo 2 v Turbo 2.5 v Multilingual?

Anytime mate! We definitely tried switching to another model - originally we had Multilingual and then tried Turbo 2.5 sadly the issues persisted, According to private forums this seems to be common for VAPI and 11labs

haha damn, that's disappointing - hopefully the VAPI team can look into that and get a solution, there wasn't really any other good (if at all?) Aussie voice models

100% man haha, fingers crossed - we currently use Savannah (VAPI voice) its ability for natural pauses and emotion has been good so far 🙂

Hey James, team is already taking a look into it but will keep you updated about it.

I have only been using Vapi as the voice provider and it happens even when there is no API calls in the workflow. It tends to happen more furher into the conversation

Hey guys, does anyone know how to fix this error? Its triggering whenever I try to open my workflow https://cdn.discordapp.com/attachments/1387668649250127923/1392352012905087006/image.png?ex=686f385c&is=686de6dc&hm=f5410587a8c6d85950c35a939d5fe5da33c717eee49d6c5f8ebf5d28c06968c8&

@james_00601 Please provide your org id

b6c29f01-8397-41e5-81a6-a5e9ed154119 🙂

Can you try switching your org update channel to 'weekly' instead of daily?

Is this for the repeating error?

It never happens on shorter context conversations. My tests on the bug where Assistants repeat themselves (even with no workflow) turned up something: I feed the conversation to date in so it can be resumed. I noticed this only happens in conversations that are fed over ~75,000 characters or so. 65,000 character transcript is no problem. 75,000+ characters and every 2nd or so sentence, they start over. Example: "With feet flat on the floor, with feet flat on the floor, place your hands place your hands behind your head with the elbows out to the sides. Place your hands behind your head with elbows out to the sides. Gently arch backward. Looking up, place your hands behind your head with elbows out to the sides. Gently arch backward, looking up slightly place your hands behind your head with elbows out to the sides. Gently arch backward. Looking up slightly. Focus on bending from the upper back, not the lower. Place your hands behind your head with elbows out to the sides. Gently arch backward, looking up slightly. Focus on bending from the upper back." Seems like a Vapi thing, not a model context thing, because switching models (the ~200k token context

anthropic/claude-sonnet-4

to the ~1M token context

google/gemini-2.5-pro

) has exactly the same character limit before it starts repeating itself every 2 sentences. Hope this helps narrow things down for the team! Takeway: It never happens on shorter context conversations.

Thank you for providing that insight. Are you also changing the maxToken value for nodes with large character count?

@Vapi @Shubham Bajaj REPETITION ISSUE - Hi Vapi, this repetition issue is a real dealbreaker, guys! We urgently need help please or some idea how to fix this please. Background: We are currently building a more complex VAPI workflow for an automotive client and run into the same "repeating issues" as the others mentioned. Using consistenly Gpt4o Cluster and Deepgram2, german in all nodes. I wasted more than 50 hours trying to get it fixed, but to no avail. As others mentioned before me, the agent randomly repeats the questions to the user 2x or 3x. Must be some kind of backend, maybe different latencies, or server latency etc. issue, but I have no switch to control it. The only switch I thought would help is to increase endpointing delay to 30-40. I even set the model temp lower to 0.1-0.3 . Latest call ID: 9fc1c8b1-1b9a-43e2-9071-63eb0f3b5a72 where it was relatively "oki-ish" (3 repititions in one node): 9fc1c8b1-1b9a-43e2-9071-63eb0f3b5a72 -> in this one it is saying (translated "Perfect,I book the appoint for you on tuesday 10 o´clock, yes?"), user confirmed two times, but it kept repeating it a third time.) - check the audio file at 1:00 minute until app. 1:42 minutes - the text transcript is incorrect, don´t look at that. You will hear it ask 2-3 times. Another call id was much much worse: cbced26a-3bf5-475e-afe4-b912e8c881df. In this one instead of repeating itself within the same node, it repeated the stuff in nodes further down the workflow again and again, (came back with date and time confirmation) multiple nodes later again & again). I am more than willing to hop on a call with you and show you live in 2-3 calls what is happening. These repetitions happen like on every other call, it gets unbearable sometimes! Any help will be highly appreciated! Thanks

Hey @Damir , did you find a solution for this? We've been having this issue for quite some time now over multiple AIs and have not been able to resolve it yet!

We noticed it happened quite often with the workflow.. and it's quite disappointing to hear it's still an issue as we do plan on going back to Workflows for the more structured bots. We've gone back to using the assistants more and barely ever (if at all) have any repetition.