https://vapi.ai/ logo
Join Discord
Powered by
Urgent: No Response on Previous Credit Deduction I...
# support
a
Hi Team, We are still waiting for a response regarding the unusual credit consumption and repeated auto reload charges issue reported earlier. There has been no update from the support side yet, while credits continue to get deducted unexpectedly from our account. This issue is now directly impacting our production environment and operations are being affected because of the continuous balance depletion. We request the team to please look into this on priority and provide: Immediate investigation Temporary resolution/workaround Refund clarification for the incorrect credit usage This is becoming critical for our production systems. Kindly resolve this ASAP or provide an urgent update. org id : 9ca74e7d-c905-4bb6-9feb-1015f062ea7a Thanks.
s
Hi, I have completed a full audit of your account (9ca74e7d-c905-4bb6-9feb-1015f062ea7a) and traced the credit consumption back to your individual call logs.
After reviewing your calls over the last 30 days, I can confirm that the billing system is functioning as intended. All credit deductions correspond to real call usage. Specifically: No runaway calls: Every call has a clear start and end time. There are no "phantom" calls or stuck processes drawing credits in the background. Real-time Metering: Vapi uses a streaming billing model. During a call, small billing events fire every few seconds to cover LLM tokens, TTS characters, and STT minutes. This is not multiple separate charges, but a continuous "meter" (like a taxi) running during your active calls. Auto-Reloads: Your Stripe auto-reloads are triggering because the consumption from these calls is dropping your balance below your set threshold.
Why the costs are higher than expected: Looking at your assistant configuration, the primary cost driver appears to be your System Prompt. It is currently several thousand words long. Because Vapi sends the full prompt + conversation history to the LLM on every turn to maintain context, a very long prompt significantly increases the "Cost per Token" for every second of the call.
Recommendations to reduce spend: Use a Knowledge Base: Instead of putting thousands of words of debt relief data/scripts directly into the System Prompt, move that information into a Vapi Knowledge Base. This will allow the assistant to only "pull" the relevant info it needs, which can reduce your LLM costs by 50–70% per call.
Monitor Cost Breakdown: In your Vapi Dashboard, click on any individual Call ID. You will see a specific breakdown showing exactly how much was spent on LLM, TTS, STT, and Transport for that specific call.
Adjust Auto-Reload Settings: If you are receiving too many small charges on your credit card, you can increase the "Auto-reload amount" in your Billing settings (e.g., reloading $50 at a time instead of $10) to reduce the transaction frequency. Regarding a refund, since the investigation shows all credits were consumed by real calls initiated on the account, we are unable to provide a refund for legitimate usage.
a
However, based on the last 30 days metrics which i have attached: -Total Spend: ~$17 - Calls: 89 - Avg Cost/Call: ~$0.19 the sudden rapid balance depletion and multiple auto reloads we experienced seemed abnormal from our side and directly impacted our production environment. We understand the explanation regarding long system prompts and streaming billing, but we would appreciate a quick 1:1 call/discussion with the technical team to better understand: - Which specific calls caused the high consumption - Token usage breakdown This will help us avoid similar issues going forward. Please let us know a suitable time for a quick discussion. https://cdn.discordapp.com/attachments/1506148200883687454/1506256776478195732/image.png?ex=6a0d9a70&is=6a0c48f0&hm=7583ae27449639b562517a2b73220965e7bf28d21cdb199279182b1f83c99cd0&
s
Hi Ankur, Thanks for sharing that breakdown - we're not able to offer 1:1 calls as our support is text-based, but I can give you the full detailed breakdown you're looking for right here.
Which specific calls caused the high consumption Your top 8 highest-cost calls were: 019db621 (gpt-4o-mini) - $0.33, 2,141,984 tokens, 26 turns 019dd0c1 (gpt-5.2) - $0.29, 336,419 tokens, 26 turns 019dbf42 (gpt-5.2) - $0.27, 293,670 tokens, 25 turns 019db179 (gpt-4o-mini) - $0.20, 1,268,781 tokens, 24 turns 019e2abe (gpt-5.2) - $0.18, 216,717 tokens, 18 turns 019dd0c8 (gpt-5.2) - $0.16, 334,987 tokens, 26 turns 019db141 (gpt-4o-mini) - $0.16, 1,022,195 tokens, 22 turns 019dbeb9 (gpt-4o-mini) - $0.15, 1,008,328 tokens, 21 turns You can look up any of these directly in your dashboard under Call Logs.
Why token numbers are so large Every time the caller speaks, Vapi sends a fresh request to the LLM that includes your full system prompt plus the entire conversation history up to that point. Your system prompt is around 3,700 tokens, so each turn adds roughly that amount on top of the growing conversation. By turn 24 you're sending around 90,000 tokens in a single request - that compounds quickly on longer calls.
The April 24 inflection point Your spend accelerated on April 24 when the assistant switched from gpt-4o-mini to gpt-5.2. Before the switch your daily LLM spend averaged $0.10–$0.38. After it went to $0.27–$0.58 on the same call volume. Total LLM cost across 30 days was around $3.64, with the remaining ~$13.36 going to transcription and voice output which scale with call minutes.
The fastest way to reduce costs Moving your knowledge base content out of the system prompt and into Vapi's Knowledge Base feature is the single biggest lever — the assistant only fetches what's relevant per turn instead of re-sending thousands of tokens every single LLM call. Also worth checking whether gpt-5.2 was switched in intentionally, as reverting to gpt-4o-mini cuts per-token costs significantly for most use cases.
Let us know if you have any follow-up questions.
a
Hi Shaunak, Thanks for your insight but what my question is can you please check the credit purchase history of 14th and 15th May it is more than $200 and simultaneously if u go and check the consumption is not more than $5 so how it happens can you please find it out and provide us with answer.
s
Hi Ankur, I'm sorry to hear about this.
We're actively looking into your account's credit purchase history and consumption for May 14th and 15th, as well as the recent $10 auto-recharge that was depleted without corresponding usage.
To help us investigate more efficiently, could you share any call IDs from that period if you have them? Even if usage appears empty on your end, call IDs would help us trace exactly where the credits went on the backend.
We'll get back to you with answers as soon as possible.
a
Hi Shaunak, I'm not able to get call Ids for that period since in vapi we can see 7 days so that's why is there any thing you want from me for this please let me know.
s
We've gone ahead and added $20 in credits to your account.
Hi Ankur, Thanks for letting us know - understood that the 7-day window limits what you can see on your end.
a
Hi Shaunak, Thanks for that. Just wanted to know few things 1. What was the root cause for this issue ? and will it be going to happen in future again? 2. We have lost around $240 credits because of this so what will be the refund process be ?
3. For chat they are asking for adding payments method and our payment card details are removed why so ?
s
Hi Ankur, Regarding the missing credits and the card being removed - could you check whether you have multiple organizations on your Vapi account? It's possible the charges may be associated with a different org than the one you're currently viewing.
You can check this by clicking the organization switcher in the top-left corner of the dashboard and seeing if there are any other orgs listed. Also, for the payment card issue, please try re-adding your card from the dashboard under Billing settings — that should restore your access. Let us know what you find.
Let us know what you find.
a
Hi Shaunak, We only have 1 organization associated with this account and could you please let us know what's the cause for this issue and the refund of the credits we lost due to this.
s
Hi Ankur, Thanks for confirming.
I've escalated this to our engineering team for a deeper investigation into the credit deductions. We'll get back to you as soon as we have answers.
a
Thanks Shaunak , Waiting for your answers!
s
Hi Ankur, Could you share a screenshot of your credit/payment history? Specifically the entries around the times when you noticed unexpected credit deductions. That will help us cross-reference with what we're seeing on our end.
a
s
Thanks! I've forwarded this to the team. Could you check your DMs when you get a chance?
Hi Ankur, We've completed our investigation into your account's credit consumption, and we've found the cause.
Your credits are being consumed by Chat API usage, not voice calls. The Calls dashboard only shows voice call usage, which is why it appeared empty - but your account has been running a high volume of chat transactions that don't appear there.
Here's the breakdown: May 14–15: 2,704 chat transactions totaling ~$182 in usage, alongside only 13 voice calls totaling ~$1.80. The chats ran continuously from early morning to evening, consuming $9–22 per hour. This is what triggered the repeated auto-reloads.
May 22–23 ($10 in 10 minutes): 111 chat transactions totaling ~$9.81, with zero voice calls. This matches the $10 depletion you reported. Why the chat costs are high: Your chat assistant is using a large number of tokens per turn (~37,000 prompt tokens), likely due to a very long system prompt or full conversation history being resent on every turn.
Each turn can cost anywhere from $0.01 to $0.75 depending on length.
What to do: Check the Logs section in your dashboard and filter by chat to see this traffic Review the assistant being used for chats - consider shortening the system prompt or moving content into a Knowledge Base to reduce token usage If this chat traffic wasn't initiated by you, check your API keys for unauthorized usage
Since the credits were consumed by legitimate chat usage on your account, we're unable to process a refund.
I'll also send you a CSV with the detailed usage breakdown via DM so you can verify the charges yourself.
PreviousNext
Powered by