I'm setting up NemoClaw locally to handle some internal documentation. I've also used Cursor's built-in AI safety features.

My main worry is telemetry. I need to understand what each system sends out when a guardrail is triggered.

For Nemo Guardrails:
* Does the "rail logs" event get sent anywhere outside my machine by default?
* If I run the NemoClaw container, does it phone home about blocked prompts?
* What's actually in the log payload?

For Cursor:
* Their docs mention "anonymous usage data." Does a safety-triggered block count as usage data?
* Is the content of the blocked prompt or its context included?

I'm trying to minimize information leakage. A local block should stay local. Which of these two setups is more transparent about what it transmits, and which one actually leaks less by default?