hi there, I would like to report repeating issue where agent talks what it does when extracting variables during the call, which are defined in Post-Call Data Extraction variable list. I would assume extraction should be done post call not during the call, right? however, I have multiple real calls where {{user_email}} was extracted during the call and client hearing it. Eg. Agent: Internal: {{user_email}} = XXXXXX@gmail.com ; here are last calls where now 2 variables where extracted with voice! Call IDs: call_b98239149670a768f41953ff3e5 call_486c24c042855a4dfd31dcb3a5c call_ef083a3086c15fcfb3b2b532647 Post call data extraction is set to use GPT 5.1 https://cdn.discordapp.com/attachments/1462714558786572339/1462714558975312016/image.png?ex=6979be96&is=69786d16&hm=af48ae11c5cadd54bb7ffb856faece02d88b2f0f06a530cb17b259606b8cbedf&

Hi @Q4

thanks 😉

Hi @Q4 This issue occurs because the input token length exceeds the LLM’s optimal limit of 22,000 tokens, which can lead to inaccurate responses. To improve accuracy, please keep the input length under 8,000 tokens.

Sorry, do you mean I should keep "Global prompt' up to 8000 tokens?. Yesterday i found that those instructions {{user_email}} and {{follow_up_desicion}} where in global prompt. Written same as they were pronounced by agent. So not sure to what you are referring about 22K and 8K tokens. If I write instructions how agent should work, I am not expecting that instuctions will be told to the caller word by word.