I am serioulsy frustrated with this platform and the bugs with it... I am trying to call an api tool that has path parameter Example [GET] https://api.test.com/{{someVar}}/resource now, this variable is passed in to the squad and squad is using an agent that uses this tool.... the damn path parameter is just not getting filled... I tested this with Agent, it works... i test it with Workflow it works... but it doesnt work with squad... Worklist is not recommend to use by Vapi... and Agent is bloated... that the hell do i do here?

Vapi is lagging on the call, agent cuts out. Is there something we need to do on our end?

Hello, as we all know, GPT 5 has been pretty bad for vapi, but how are the two new 5.1 models? Are they any good? I'm also curious to know if the latency times are correct, 5.1 instant is just as fast as plain 5.1. It also has the exact same specs. Any noticable difference there?

I've previously been using 'experimental' warm transfer but it doesnt work 100% of the time. So I'm trying the TWIML method. I have it working speaking a message "this is a call from company, the customer is called..." and at this point I want it to say the customer name. I've used {{customer.name}} but it reads out "customer name". Is this possible and if so how? The same with call transcript/summary. On the other call transfer methods the options to create the summary plan are there, but it doesn't show on the TWIML configuration.

I'm trying to figure out why, for the transcript, the LLM output is sent to the speech-to-text. Why not just directly send me the LLM output? Is this a setting I can configure? We're building out a third-party web widget that's pulling in the transcript, and it's taking it from the speech-to-text. It would be better to just give us the exact LLM output, and I can't figure out how to just get that exactly, the actual LLM output. So I could put that into my transcript for the user.

I am using the VAPI Dashboard on MACBOOK in Browser CHATGPT ATLAS and it is lagging like crazy.

Everybody's seen all of the post. With this error I've tried everything that you've mentioned. Created the new API keys, cleared the cache, did a hard reset. Still nothing. I see multiple people with this error and it's gone unresolved. Can anybody help here?

Is it possible to update the assistant system prompt during a call? Do dynamic variables get updated every turn of the conversation? Thanks!

Hello! I'm following the Live Call Control docs at https://docs.vapi.ai/calls/call-features , but I don't see any monitor object being returned after I create the call. Am I missing something? Or perhaps I'm not on the right Vapi tier?

Details: I am creating a recruitment bot using Gemini-Flash 2.5 and VAPI. During the conversation, it keeps breaking with the error shown in the attachment. It has been two days and I still haven’t been able to resolve it. With Gemini-Flash 2.0, it works well. Call ID: 019abb65-1aa5-788e-b615-f786b2e7a538 time: 1764081022769 Response Expected: Next set of question Error: Please see attached. Console error is pasted below: could not emit {action: 'error', errorMsg: 'Meeting has ended', error: {…}, callClientId: '17640821643690.5208472250220907'}action: "error"callClientId: "17640821643690.5208472250220907"error: details: undefinedmsg: "Meeting has ended"type: "ejected"[[Prototype]]: ObjecterrorMsg: "Meeting has ended"[[Prototype]]: Object Error: Unhandled error. (undefined) at Y.emit (index.js:38:16915) at gu.emit (index.js:38:44170) at r. (index.js:38:48219) at i.emit (index.js:2:4192) at r.value (index.js:30:33247) at r.value (index.js:30:27805) at o (index.js:22:16504) https://cdn.discordapp.com/attachments/1442895592731705437/1442895593084031078/vapi.png?ex=692718bd&is=6925c73d&hm=4d13dd553594e131de6b53bb40f7a5d91168ad7a490e2239ed15931fdd785e7a&

Hey everyone, I have connected a Twilio number in Vapi and am receiving incoming calls, but the call log still shows the inbound call with Assistant = “No Assistant Assigned” even though the assistant is set in the phone number settings. Info: Org: Twilio number: +41 43 505 10 54 Assistant name: Demo Inbound settings in the dashboard: Inbound phone number = +41435051054, Assistant = Demo Twilio: “A call comes in” → Webhook https://api.vapi.ai/twilio/inbound_call (HTTP status 201 in the Twilio call log) In Vapi calls, I see my call as Type = Inbound, but the ASSISTANT column says “No Assistant Assigned.” When I test the same assistant directly via “Web Call” in the dashboard, everything works (call log then correctly shows Assistant = Demo). Can you please check whether there are multiple PhoneNumber objects for my Twilio number, or why the inbound handler does not link the number to the assistant even though it is set in the inbound settings? Relevant call IDs from your log: 019a87ae-ea68-7773-aa33-fc1adeafe44b Thank you & best regards, Silvio https://cdn.discordapp.com/attachments/1442893938225713234/1442893938565578772/4a986c99-908c-44d4-97f8-51673d6c4493.png?ex=69271732&is=6925c5b2&hm=60e978bdb50d9ddffafde2e12f4f651ddb546e369f8da8bf6db4eefb97d50142&

We're using elevenlabs as primary voice provider, but when we change the synthesis settings, is not reflecting. We figure out that this is a caching. Which is not a problem at all. On other hands, we would like to know 1. What is the TTL for cached voices? 2. Is there any way to enfore caching expiration? PS: We don't want to disable the caching at all

I'm trying to integrate vapi to my cal.com calendar to book appointments with clients automatically, but vapi is booking at unavailable times and the info is not being transferred to the calendar. can you help?

I already created a post with this https://discord.com/channels/1211482211119796234/1435697639743033436 but it seems after a while posts don't get any other answer so I am reposting to maybe have it fresh and get some eyes on this. I already tried the configuration suggested and it does not work. It does not even allow to set that configuration.

I have a Vapi assistant in my dashboard, and I want to connect it to three different bot interfaces: mobile, web app, and phone calls. For each connection, I also want to pass the customer’s details and name using the assistant override feature. Can this be done, and if so, how? Additionally, I want to enable call-forwarding only for users who contact the assistant via phone calls, and not for those connecting through mobile or web applications. How can I configure this behavior?

I'm using a SIP trunk. Outbound worked, but inbound didn't. So, the system isn't picking up incoming calls. How can I fix this?

Hi everyone, how can I send a demo link for people to test my squad of assistants? Normally, for individual assistants, I can do this, but for squads, I can only share a phone number right now. https://cdn.discordapp.com/attachments/1442812696855711754/1442812697178804235/image.png?ex=6926cb89&is=69257a09&hm=f13279bee3b494b572e309553a28ce27eb7dfd9fc18fa207a37975752b660f98&

I'd request your urgent help to this matter, our systems are scaling fast (and in Production) and Vapi APIs are responding in a very unreliable manner. Hi Team, We have a system with following properties 1. We have got 5 numbers where each number handles one call at a time. (We have limitation of one active call at a time from number provider) 2. A number is marked busy when a call request is sent to Vapi 3. A number is freed up and ready to accept calls again when we receive status ended from Vapi Webhook We still receive concurrency limits errors a lot of time. (even though we should never considering Vapi default concurrency limit is 10) For an instance, On 24 Nov 2025, at 4:43 PM - 4:44 PM (IST), there are 10 calls which are showing in Vapi call logs which were all customer busy or customer did not answer Still, all the calls between 4:44 PM - 4:48 PM failed with concurrency limits hit error. (there are no call logs between this duration in Vapi signifying that system was free during this time) It seems that Vapi doesn't free up concurrency sometimes even when there are no active calls going on. Thanks, Karun

Hello, I'm interested in connecting a voice agent for up to 250 lines. Who can I speak with?

We are currently evaluating Vapi as the orchestration layer for a healthcare voice agent application. We understand there is a $1,000/month add-on for HIPAA compliance and Zero Data Retention (ZDR) on the Vapi platform. Before we commit, we need definitive clarification on how this add-on interacts with downstream model providers, specifically regarding Business Associate Agreements (BAA). We have been unable to definitively clarify from the help docs here: https://docs.vapi.ai/security-and-privacy/hipaa Could you please clarify the liability in the following two scenarios? *Scenario A: Vapi Managed Keys * If we pay the $1,000/mo HIPAA fee and use Vapi's default/managed keys for Deepgram (STT), Anthropic (LLM), and Cartesia (TTS): - Does Vapi's BAA with us cover these downstream providers as your sub-processors, and are the sub-processors required to adhere to ZDR? - Or are we still required to hold our own separate enterprise contracts/BAAs with Deepgram/Anthropic/Cartesia to ensure they do not retain PHI or violate ZDR? Scenario B: Bring Your Own Key (BYOK) If we bring our own API keys for these providers to get direct pricing: - Does the "HIPAA Enabled" toggle in the Vapi dashboard technically prevent these downstream providers from logging data on their end? - Or is the ZDR status strictly dependent on the settings/contract associated with our specific API keys? We are trying to determine if the $1,000 fee provides a "compliance wrapper" for the entire stack, or if it only covers Vapi's orchestration logs.

Hello, I’ve been trying to get in contact with someone from Enterprise Sales. I need to discuss signing a DPA. Could someone please reach out to me?

Hi there, we have been using a twilio telephony number with transfers and noticed our twilio is paying for the call after the transfer. We are switching over to a SIP trunk connected number on twilio and will add transfer blind transfer with verb REFER, but i am curious whether if we transfer to a sip system (but through a phone number) will their system take over the call?

It's all set up and with a system prompt that takes into consideration the proper numeric pronunciation and style to present the information in the Spanish language. Strict direction to speak out in Spanish I see the 11Labs and Azure and others 'supposed best voice interfaces' to fail and mixing up the numbers and some wordings with a strange intermittent language and sometimes english and french. $20.57 is 'veinte (correct)' with fifty seven (incorrect) cents. I tried the most powerful models such as Llama 3.1 & 3.3, came back to GPT 4.1 but nothing... I'd like to line up a troubleshooting ticket with 11Labs in parallel but I don't want to raise a ticket with 10 service providers, it all should be converging here. BTW, using Nova 3 and it works fine (spanish driven)

message above says it all orgid: 14d31e07-98c4-4f99-aefd-4d35ed5baa80 https://cdn.discordapp.com/attachments/1442572041285796061/1442572041554362522/image.png?ex=6925eb68&is=692499e8&hm=ccbb00174b20fa7f09745556daa9cd4a8fca908d9bc35ff9bc56b5a16f03b098&

I accidentally made one of my developers an Admin who no longer works with us. And when I tried to demote him, I ended up demoting myself. Now I can't access the admin portal. I have no idea how I can fix it. My credit card has been added to this account. I need help to return the access as soon as possible. Someone, please help me

I am trying to find the VAPI cost per minute for 4o model and GPT5 model. How can I find this? I need VAPI cost for both models so i can decide what to offer my customers.

We send control messages mid-call to let the agent know about new information. We're expecting to see this message in the "conversations" field in any mid call webhook. We do see this message, however it is often truncated which is causing issues on our system. We send a json as this control message and when it's truncated we lose information. Could we stop this truncation? Ex:

{
        "role": "user",
        "content": "I sent it."
      },
      {
        "role": "system",
        "content": "[{\"type\":\"user-send-email\" [ some more characters] ..."
      },
      {
        "role": "user",
        "content": "Have you gotten it yet?"
      },

Hi I'm working on a platform for a client where I do calls for them using the vapi voice agents. The numbers are imported from telnyx through sip trunking. There are multiple issues that need to be resolved. 1. One of them being the latency, it's sometimes around 4-5 seconds, and most people end the call thinking that there's no one responding. 2. Inbound calls seems to be not working. 90% of them end after agent greets the receiver, and yet the call status shows that customer ended the call. So it's confusing what's going wrong 3. In some calls agent keeps speaking but it seems like customer doesn't hear it as they keep saying hello. Some of the call ids: 019a956e-285b-7003-90a0-65a177cc900b 019a9527-2ecd-711a-a40d-d76e31331268 019a951f-ae4f-7229-a61c-a7426dcf94b4 019a951c-ac9a-7113-80b6-68d4291d1930 019a94ff-cb1a-7aa3-88f8-49b89fcc3cd7 019a94f4-a22c-7881-9ab0-986c4ed2ac43 019a94ed-6669-7551-acce-5ddc38894aab 019a7115-f015-7995-9fb0-ab927b8479bd Need to get this resolved as it affects my clients badly

We recently created a sip trunk and a phone number id: ed8ae47a-1797-4d51-ade5-472736461902 and also assigned the phone number to the agent, but we keep getting these errors. Essentially the invite is being sent by our provider, but they say that vapi isnt accepting them. https://cdn.discordapp.com/attachments/1442520644733636698/1442520645136158760/c523ddc1-5b15-44e7-9a57-53fecee7239d.png?ex=6925bb8a&is=69246a0a&hm=a65a6b6da8ccea9e4703709b0be2845c171fb43ed30aa2021ae068c705b42b07& https://cdn.discordapp.com/attachments/1442520644733636698/1442520645945528351/f5f6e43c-6d74-44f3-9466-7ca9f0220259.png?ex=6925bb8a&is=69246a0a&hm=f38a60b8eb98a6900e1fbfd3b9e89614cd46b62a1e86e908e39ea7cbe438ba84&

I want to adjust my prompt in case the caller called the agent or if I initiated an outbound call. Is it possible to do in a similar way to {% if transport.conversationType == "chat" -%}