Agent OS is a kernel architecture designed to manage autonomous AI agents with a focus on safety and policy enforcement. It allows developers to create agents that operate under strict rules, ensuring actions like database queries adhere to specified policies. For example, if a query attempts to execute a destructive command like "DROP TABLE," the kernel blocks it based on the defined policy, which enhances security and reliability. The system provides a straightforward setup. Developers can get started by installing the Agent OS package and using just a few lines of code to create a governed agent. The StatelessKernel class allows for immediate policy enforcement, so agents execute safe queries while blocking harmful ones. This level of enforcement is achieved without relying on the AI's compliance with prompts, as the kernel itself evaluates actions beforehand, similar to how traditional operating systems manage permissions. Agent OS employs a layered architecture, separating concerns into distinct modules for better organization and functionality. Key components include the Policy Engine, which checks actions against set rules; the Flight Recorder, which logs agent activity; and support for memory management through the Episodic Memory Kernel. This modular approach allows for scalability and integration with various frameworks, making it easier to build complex agents while maintaining safety and governance standards. Various examples and demos are available to illustrate the platform's capabilities, including SQL agents and compliance bots.