Adrian

What it does

Adrian is a runtime security monitoring and control system designed specifically for AI agents. It operates in real time to catch and prevent security threats before they can cause harm, including malicious tool usage, prompt injection attacks, and unintended policy drift. The tool provides continuous observability into agent behavior and enforces security policies dynamically.

How to set up

Adrian integrates with LangChain, LangGraph, and MCP-compatible systems. Installation typically involves: 1) Adding Adrian as a dependency to your agent framework, 2) Configuring security policies and monitoring rules, 3) Initializing the runtime security layer in your agent execution pipeline, 4) Setting up alerting and logging for security events.

Key features

• Real-time detection of prompt injection attacks
• Malicious tool use prevention and monitoring
• Policy drift detection and enforcement
• Integration with LangChain and LangGraph agents
• MCP protocol support for compatibility
• Comprehensive observability and threat detection
• Pre-execution action blocking capabilities