From what you’re seeing, the Server URL in the “Messaging” section is mostly for events and status updates, so Vapi AI will still fall back to whatever is set in the dashboard. That’s why your custom firstMessage and system prompt are being ignored. For real overrides to work, your backend needs to be connected to the assistant/voice request flow, not just messaging. Also, using assistant-speaks-first-with-model-generated-message can sometimes override what you send from the server. Are you returning your dynamic config from the actual assistant-request endpoint, or only from the messaging webhook?