Adrian is an open-source runtime security monitoring and control tool built specifically for AI agents, published in the secureagentics repository on GitHub.
Its purpose is detecting and controlling malicious or misaligned agent behaviour while the agent is actually running, rather than reviewing what happened afterwards.
It monitors an agent's activity logs, tool calls, outputs, actions and reasoning traces. Scrutinising those traces is what allows it to identify deviant behaviour and respond in real time, in some cases before the agent acts at all.
The distinguishing capability is optional in-flight intervention: behaviour falling outside the expected range can be stopped rather than merely logged.
For developers it is deliberately low-friction. A Python SDK handles installation, and integration with LangChain agents takes two lines of code. Adrian is free and open source.
Research basis: This is a source-based directory listing prepared from public product information; it is not represented as a hands-on test. Last updated September 20, 2026. Read our methodology.
Exact feature availability by plan was not specified in the source.
Featured on
TaskBoosters
