Me and my team were doing some extensive testing today, and noticed something very alarming. The assistant is using data from previous calls. One example here: 13c0bf3f-7a3e-48da-99ec-09c1496e7866 I did mentioned in a previous call that I wanted 55 BLT's sandwich, but not on that call. And out of nowhere it replied with this: "Here's your order. 55 BLT sandwiches with no onions, 1 large fry, everything correct?" This only happens when using the

openai

provider. I did switch to custom-llm targeting a local api that it's basically a proxy to openai and it did work correctly. Is this a bug on Vapi side?

@Nomadiction I'm sorry to hear that you're experiencing issues with data persistence across calls. It sounds like there might be an unintended retention of context or session data when using the

openai

provider. This is certainly not the expected behavior, as each call should be treated as a separate instance unless explicitly designed to maintain context. To address this issue, I would recommend reviewing the configuration settings for the

openai

provider to ensure that session management is handled correctly. It's possible that there may be a bug or misconfiguration causing this behavior. If you've already checked the configurations and the issue persists, it would be best to report this to our support team with the specific details of the calls and the unexpected behavior you're observing. They will be able to investigate further and provide a resolution. Would you like me to assist you in reporting this issue to our support team?

Just as additional context, everyone was calling from the same number since we use Telzio for making test phone calls

@Nomadiction set the semantic cache to false for your assistant and then it will work. https://docs.vapi.ai/api-reference/assistants/update-assistant