Adrian

Adrian is the first open-source security monitoring toolkit that analyses not just what your agent does, but what it thinks. Based on research from OpenAI and DeepMind, Adrian installs with just 2 lines of code and applies security monitoring to your AI agents. When your agent starts looping, gets compromised, becomes misaligned or generally starts misbehaving you get notified on Slack or Discord. Adrian is free, forever, and even supports running entirely locally on your system.