Yes, but how. Am I looking for a server message vapi sends as the trigger? How does it work

you’re basically looking to intercept the message before it hits the LLM. In VAPI, you don’t need a “server message” from the agent itself; instead, you handle it on your backend or webhook. The flow looks like this: the user message hits your webhook first, you query Qdrant for the relevant context, merge that context into the message (system or user input), and then send the enriched message to VAPI as the agent input. This way, the LLM always sees the full context, and the RAG step is fully decoupled from the agent tools. I can help you set up a clean webhook/middleware structure so this happens automatically for every turn.

This doesn't really make sense. If I don't have any message sent to the webhook, how does it trigger? Maybe it's also a client message, not sure but I need some sort of trigger for the webhook. Which one? @User https://cdn.discordapp.com/attachments/1468550381536678046/1468551369987325962/image.png?ex=69846e8a&is=69831d0a&hm=74ad75ce7731dcf47ea709e1a15d17750abd0e21b8cc77588da961a17e4e57af&

Conversation-update is probably your main webhook trigger that fires with each conversation turn/update, sending relevant data to your configured webhook URL. Without more context about what system this is (appears to be some voice/call API), I'd recommend checking the documentation for the specific webhook event types. If this is for a real-time conversation API, conversation-update would be the standard trigger for continuous updates during an active session.