Credit
how does credit consuption works? 1 credit = 1 user message? Yesterday I tested just one API block but over 12000 credits were consumed. Could you help me understand why this happened? If API blocks dont consume credits I’d like to request a refund for the credits used.
9 Replies
Hi there! Here’s a quick recap of how credits are counted and why you might see a big spike.
🔹 What costs credits in a chat agent
• 1 user-sent chat message = 1 credit
• Agent messages, API blocks, function blocks, logic steps = 0 credits
• LLM calls (GPT-4, Claude, etc.) = credits based on tokens used (see pricing table). So if you only clicked “Run” once and typed one message, that would normally be just 1 credit plus whatever the LLM spent to generate answers. 🔹 Why 12 000 credits could appear 12 000 credits ≈ 600 USD of usage. The two most common causes are: 1. A loop in your flow that repeatedly hit an LLM prompt.
Example: GPT-4-turbo (2.5 credits / 1 k tokens). 12 000 credits ≈ 4.8 M tokens – that can happen fast if a loop runs for minutes.
2. You accidentally ran a voice test – voice minutes are 10 credits each, so a 20-min call = 200 credits. 🔹 How to verify 1. Open the project ➜ Analytics ➜ “Credits usage” chart.
2. Expand the day in question – you’ll see a breakdown by feature (chat, voice) and by model (GPT-4, GPT-4o-mini, Claude, etc.).
3. Cross-check which step in your flow fired the model/API repeatedly. 🔹 Next steps • If you find a loop, add exit conditions or move the LLM prompt outside the loop.
• To avoid surprises while testing, use cheaper models (GPT-3.5, Gemini Flash) or switch LLM blocks off until the logic is stable.
• You can also set hard credit limits for the workspace to stop the agent when the limit is reached. Docs for reference: • Credits overview – https://docs.voiceflow.com/docs/credits
• Pricing table – https://docs.voiceflow.com/docs/credits-pricing-table Hope this clarifies what happened! Once you’ve looked at the analytics breakdown you’ll know exactly which block consumed the credits and can adjust accordingly.
• Agent messages, API blocks, function blocks, logic steps = 0 credits
• LLM calls (GPT-4, Claude, etc.) = credits based on tokens used (see pricing table). So if you only clicked “Run” once and typed one message, that would normally be just 1 credit plus whatever the LLM spent to generate answers. 🔹 Why 12 000 credits could appear 12 000 credits ≈ 600 USD of usage. The two most common causes are: 1. A loop in your flow that repeatedly hit an LLM prompt.
Example: GPT-4-turbo (2.5 credits / 1 k tokens). 12 000 credits ≈ 4.8 M tokens – that can happen fast if a loop runs for minutes.
2. You accidentally ran a voice test – voice minutes are 10 credits each, so a 20-min call = 200 credits. 🔹 How to verify 1. Open the project ➜ Analytics ➜ “Credits usage” chart.
2. Expand the day in question – you’ll see a breakdown by feature (chat, voice) and by model (GPT-4, GPT-4o-mini, Claude, etc.).
3. Cross-check which step in your flow fired the model/API repeatedly. 🔹 Next steps • If you find a loop, add exit conditions or move the LLM prompt outside the loop.
• To avoid surprises while testing, use cheaper models (GPT-3.5, Gemini Flash) or switch LLM blocks off until the logic is stable.
• You can also set hard credit limits for the workspace to stop the agent when the limit is reached. Docs for reference: • Credits overview – https://docs.voiceflow.com/docs/credits
• Pricing table – https://docs.voiceflow.com/docs/credits-pricing-table Hope this clarifies what happened! Once you’ve looked at the analytics breakdown you’ll know exactly which block consumed the credits and can adjust accordingly.
Hey @matcastro, could you share more details here? That is highly unlikely, that would be nearly $60 of OpenAI usage for a single call
Are you looking at tokens in the debug logs?
I had to delete the agent so my credits would not be consumed
And the only block that i had is an api call
Got it, do you have any screenshots or anything? Hard to go off this and see what happened, it's just super unlikely so would want to see what happeend
can you share your VF email? our support team will pull the logs and see what happened
chatfy.vf2+2@gmail.com
I dont have any screenshots bacause i had to delete the agent very quickly
Thanks very much braden
what project is this....
i think this is the most agents i've ever seen haha :p 1500+ agents
honestly it might not be 1 API call, but looking at the totals - they aren't real-time and update every few minutes
@matcastro
The project was Eccomerce - Recuperação de Vendas (COPY)
But was deleted
Sunday night
Got it
hm
I'll DM you
Ok thankss