question, on the VAPI website it says the cost for LLM (4o) is $0.07 per 1M tokens while open charges $2.50 for the same 1M tokens, does this mean VAPI is subsidizing the LLM cost ?

Dear Vapi.ai Support Team, I am writing to report a CRITICAL PRODUCTION ISSUE that is making our AI voice agents completely unusable for customer-facing calls. ISSUE SUMMARY: Our AI assistants are literally speaking variable syntax like "{{name}}" to customers instead of processing dynamic variables when values are null/undefined. ORGANIZATION DETAILS: Org ID: 9e2fba01-3a3e-44b6-b51d-da0bf706f586 Call ID (Example): 1e66d015-d64d-4fce-8478-25fd7384eeda ACTUAL CUSTOMER CONVERSATION: AI: "Hi {{name}}, thank you for calling Light Lounge Holland my name is happy. How may I help you?" Customer: [responds] AI: "Perfect, thank you for confirming! I have your name as {{name}}. Is that correct?" Customer: "What do you have my name as?"This is happening to LIVE CUSTOMERS calling our business. TECHNICAL DETAILS: No changes were made to our system - this worked correctly before Issue occurs when dynamic variables ({{name}}, {{email}}, {{phone}}) are null/undefined Variables are passed correctly in variable_values during assistant requests System prompt clearly handles null values, but variables aren't being processed at all BUSINESS IMPACT: PRODUCTION DOWN: Cannot use AI for customer calls CUSTOMER EXPERIENCE: Extremely unprofessional interactions REVENUE LOSS: Customers hanging up due to broken AI experience BRAND DAMAGE: Makes our AI system appear completely broken EXPECTED BEHAVIOR: When {{name}} is null, the AI should either: Skip the variable entirely Follow system prompt instructions for missing data NOT speak literal "{{name}}" to customers IMMEDIATE ACTION REQUIRED: This is a production-critical bug affecting live customer interactions. We need: Immediate hotfix or workaround Root cause analysis of variable processing changes ETA for permanent resolution URGENCY: This issue is preventing us from serving customers and causing significant business disruption. Please escalate this ticket to your highest priority level.

Call ID: 5c5a2da1-8051-42b2-833a-5d1dc5a4 Expected: Transfer should have been successful Actual: Transfer failed. For the caller, it just played 3 tones, and ended the call. It seems that the where assistant is hosted in your server cluster, it is hitting some server or network level error which is preventing it from call transfer to Twilio. Same call transfer tool was working with another Assistant. I compared all the settings but no luck. I just duplicated the the other assistant, and updated its prompts, and it started working.

A person keeps filling in our website form to make our voice ai call his numbers. He keeps doing this with multiple numbers around the world and repeating the same sentence again and again just to drain all our credits. What can I do? Can't block as he keeps using new numbers

hi everyone, i have an agent setup and everything is working well except getting it to transfer a call out to a human. i have created a tool and then tried a workflow and in both cases, the agent says it's transferring the call and then just hangs there. any suggestions?

@Vapi Support Bot @User how to use call listenUrl in browser (react app) to listen to the live stream?

I'm trying to log in and it tells me that I need to do a security validation. I did it and it doesn't log me in.

I recently had a problem with a developer. I removed him. Shortly after, I am no longer able to log in. When I verify my identity, it says successful, but then it just hangs while "verifying" on vapi. Long after the 1 to 3 minute window. I have tried this multiple times over several days and emailed support. My concern is that he intentionally sabotaged my account, or that there are charges racking up. Lost password simply generates an error.

Hi, i just discussed this issue during Arthur's OH, and he suggest to raise a ticket as a debug is needed. We set the waitSeconds to 3.0 seconds for the entire call, but in some cases the BOT start to speak while the user is still speaking. I believe it happens when the user start to speak, stop to speak for a second, and start to speak again. the waitSeconds timer maybe start the 3 seconds count from the brake time and interrupt the user while he is still speaking. See call ID a207f0d6-8403-4c07-a650-a75ef8c9ba52 as an example. time 16:05:00:347 as a good example. another time example inside this call: 16:05:28:359 see attached screen shot with this call graph. i marked the 2 example above and another one. You can see the bot started to speak immediately after the user (or even before the user ended on the second example), and you can see the user speak with a short brakes in the middle which probably cause my problem. i would expect VAPI to reset the 3 seconds timer once the user start to speak again. Let me know if you want me to try and recreate such a scenario myself while speaking in English. Hopefully you have enough details to debug with this call ID. https://cdn.discordapp.com/attachments/1389323129368612875/1389323129620267038/Screenshot_2025-06-30_220936.jpg?ex=6864337e&is=6862e1fe&hm=fe6d3d49b83c38e0c1108dda8eeaa2c52dfbc8168074fbed4b38bcfcdf40b34b&

^

I know you can do structured data outputs from assistants, but I was wondering if we can do a similar thing in workflow calls.

I run a software dev agency and I’m evaluating different voice ai options for their call center. looking to get enterprise pricing for 165k minutes per month, haven’t heard back from my email

Good afternoon, I have the following case. In the VAPI platform, under the Billing section, there is a chart showing the number of minutes per day. However, this information does not match the number of minutes I extracted using the VAPI API. I’m attaching the file and a chart with the extracted call data. For each call, if there was no cost, I set the minutes to zero. If there was a charge, I included the number of minutes corresponding to the cost. As a validation step, I calculated the cost-per-minute ratio and got exactly $0.18 USD/min, which matches VAPI's pricing—so the data seems consistent on my end. I would appreciate your help in understanding and clarifying this discrepancy between the dashboard and the API data. Thank you in advance. https://cdn.discordapp.com/attachments/1389294036245807104/1389294036510183425/hsbGxibDDhgbGvrG2hr66iqqsnJyYuKigsKCgMAMUlTiHn9lZYAAAAASUVORK5CYII.png?ex=68641866&is=6862c6e6&hm=69c573a09393fbcc8503adde3449399571bafc228c98835b19e9e980cdad5f5c& https://cdn.discordapp.com/attachments/1389294036245807104/1389294036791197859/VAPI_cost_and_minutes_chart.png?ex=68641866&is=6862c6e6&hm=0cb4fda4845268b41ab745676189dd9ed66239b047323d6731f448c0121aba70& https://cdn.discordapp.com/attachments/1389294036245807104/1389294037038534779/vapi_api_calls_out.csv?ex=68641866&is=6862c6e6&hm=3951f286cf5b2e3820e82535e2874c9ab8b2aa02705cdd4794f8636cd0fd798a&

Other members are saying it's active and I have admin access, but I can't access it. Haven't been able to for weeks. Tried resetting browsers/computers, updating things, etc. Org ID: 9171eae7-9f64-462c-8ea2-493cbe50f00f Email: zack@thinkbullish.com

Hi there - New user, got the 1000 credits free but now when I try to pay with debit, credit cards nothing works. HELP!

I have a UK number from Zadarma, I have connected it to VAPI (added the SIP Trunk Credentials and imported the phone number). When I tested it calling a Greek number from this phone it worked as expected. (example call ID: bc24e9bd-6423-4e6b-89a0-c68d5c5f7734) When I made a call to UK numbers it always ends receiving this automated message from user: "These calls are only possible within the country." (example call ID: bf70d462-c62c-4e0f-9021-594cbc98c163) How can I fix this?

As the title suggests, When I try to save the secret token with any assistant, it will not publish, though the pop uo says it will. Happy to provide assistant ID and video recording in a DM if it helps

When a caller asks if we are open on a date and time the agent is not able to determine the date and time how do I fix this?

I'm trying to make it so in my Vapi workflow I can play an audio clip. I tried setting up a Vercel api but it keeps giving me a network error when I test the tool. Any thoughts?

Hi everyone, I’m having an issue with my AI call assistant. The system is supposed to say this sentence to the caller before transferring the call: “Alright, I’ll now transfer you to the nearest available expert who can help.” But instead, the AI skips this message completely and forwards the call straight away — sometimes saying only “Transfer the call now,” which isn’t part of the real conversation with the caller. I’ve made it clear in the system prompt that the AI must say the transfer message out loud to the caller before triggering the transferCall function, but it keeps ignoring that instruction. Can you please help me fix this? I need the AI to always say the transfer message to the caller before the call is forwarded.

This is the call id 28c1f844-bb8c-489d-a5bc-4a11e0605140 In the call bot speaks with low volume and then suddenly it increases the volume. Can you please explain why is this happening? Requirement: We need a NZ accent male voice, without the above mentioned issues. I have also used PlayHT's voice, it also has issues.

While trying to sign up with my new company mail it says Confirmation mail sent, but i got nothing in my mail. email : amir.z@marc-ellis.com

I have several vapi accounts where i can't get through this step. please help! https://cdn.discordapp.com/attachments/1389246326067302541/1389246326352511058/image.png?ex=6863ebf7&is=68629a77&hm=10aee8c16f7785915224fb84dc76a022265e01d5f6026bb5713037a5c570c8dc&

So there's this weird issue going on with vapi workflows, whenever there is a tool call used, and it fetches a response, the assitant always says the response twice. Like "I found the available times, Here are the times ...." "I found the available times, Here are the times ...." Repeating the same thing twice, Anyone know what can cause this? Call ID: 9954a7c5-bbe2-4b3b-a245-12d9d9f7c975 Timestamp: 0:40

Im using Make.com in order to make outgoing calls with my assistant, seems that transcript is missing on the endcall report and im getting Summary as “I apologize, but the transcript of the call was not provided. please provide the transcript so I can summarize the call for you”. What can I do?

I am not sure what the issue is but we have not been able to login to our vapi account and have gotten the following emails. Have reached out on email too but no response. For the record we have a single account that is accessed by all devs who do live in different regions which could be a reason for logins from various locations. Please resolve this ASAP as we have clients that get affected through this @Vapi Support Bot @Vapi @VAPI Tech @Shubham Bajaj https://cdn.discordapp.com/attachments/1389178349250809967/1389178351607873557/vapi_blocking.png?ex=6863aca8&is=68625b28&hm=cbbe599ae434f66d920ba50612da36676fd90bfb56944871043d81cd3e52c002&

When I receive a webhook with type: assistant-request i am preparing my responce and sending it back to the workflowId (specific workflow).

{
  "workflowId": "this-is-my-workflow-id",
  "workflowOverrides": {
    "variableValues": {
      "first_name": "Peter"
    }
  }
}

at the workflow i add this varuable {{first_name}} at the prompt and at the first message of the conversation node, and when call (inbound) my phone number, the agent did not see the value of the variable. * The phone number is imported and it is not assigned to a specific workflow. According to the documentation: https://docs.vapi.ai/api-reference/webhooks/server-message#response.body.messageResponse.AssistantRequest.workflowOverrides.variableValues this is the right way to respond to the webhook in order to override the variables at the Workflow. Please for your help because I am stucked with this and can not continue. My workarround is to start with an api call first and get the data and then start the conversation but this is not ideal for me.

Hi i want to know if there is a way, we can connect the voice agent from our server to custom javascript web socket or api's?

I have an API request that takes in one parameter; date in yyyy-mm-dd format, the "Assistant" bot seems to handle this find but my "Workflow" bots seem to struggle putting dates in the correct format. Is there a sure-fire way to guarantee it's always using yyyy-mm-dd format? Things i've tried: - Added to Global Prompt - Added instruction to use yyyy-mm-dd in the api call description - Added instruction to use yyyy-mm-dd in the conversation node Tool that works with Assistant: 1ab710d2-c53c-48e7-a06e-53c25dc64dbd Worfklow that doesn't work: 34fc1cf5-eb87-4d3c-9e96-4189e3222753

is the 10 concurrent call limit across the phone numbers or every phone number can have 10 concurrent calls, (say if we have 10 number thent total calls will be 100)