Hi team, we are noticing our AI assistant is not maintaining context that well. Here is the conversation: User: Hi, Jasmine. I'm coming to the club tonight for dinner, but I'd like to put the kids into Play Area. Can you just let me know what time it opens? AI: Kids play area opens tonight 4 30 PM and will be open until 8 30 PM. If you're bringing the kids, just a quick reminder that it's on a first come, first served basis, so it's best to arrive early. Is there anything else I can help you with? User: Okay. And are the opening times different for the weekend? AI: Yes. The opening times are different on weekends. For both Drinks Bar and Sports Bar, they are open for dining from 11 AM to 11 PM, on Friday and Saturday. You can see above that our Assistant switched from talking about the kids play area to the bars. How can we ensure that our assistant maintains context in conversations? Is there a setting in Vapi?

what model are you using? this is not a vapi setting. this is your prompting and model choice.

Hi @Bryce , we switched from Nova 2 to Nova 3 and have also added prompts. Do you have any suggestions to improve maintaining context?

What AI model?

@Bryce AI Model - GTP 4o mini Using Trieve for knowledge base

Hey Beck To help track down this issue, could you share: - The call ID - When exactly this happened (the timestamp) - What response you expected to get - What response you actually got instead This would really help us figure out what went wrong!

There ya go. Don’t use 4o mini it’s trash. Use 4o. The extra 150ms of latency is worth it

Thank you so much @Bryce, this has solved so many issues for us. We appreciate it!

Marking this ticket as solved ✅

Hi all, I am hoping to find a quick solution for this, no idea if it's to do with Maintaining Context. This assistant's task is to collect customer information for functions. She has 8 pre-qualified questions/info to collect. When there is some kind of interruption (e.g background noise) or a subject change (e.g custom asks a different question), she does not go back to complete the pre-qualified questions (information is not asked and collected) and she does not trigger to send email. For this specific call below, we also have result as "Success" but it was not as it did not complete the task: { "name": "endCall", "role": "tool_call_result", "time": 1740112040461, "result": "Success.", "toolCallId": "call_obuZtghCpRIHPLq6hicfKQzn", "secondsFromStart": 198.21 } CALL ID: feff38ad-9a15-4a35-96b6-a60f39bde87d Timestamp: 02:58 What we expect to happen: Assistant goes back and continues the Pre-Qualified questions, ends the call and fires off email trigger What happened: Assistant "forgets" where she was at, starts a whole new conversation "Hello, thanks for calling, how can I help you?", says she called at the wrong time (when she didnt make the call as it was inbound), ends the call abruptly and does not fires off email trigger.

@firefly_0206 please create a new support ticket.