Show HN: I built a firewall for agents because prompt engineering isn't security (github.com)

Cordum has unveiled a new control plane for autonomous AI agents, operationalized as a "sudo" mechanism that facilitates fine-grained control over workflows and decision-making processes. Utilizing technologies such as NATS for messaging, Redis for state management, and CAP v2 protocols for job handling, Cordum is designed to enable deterministic automation with built-in policy enforcement mechanisms. This framework supports features like least-loaded scheduling, approval gates, and a robust workflow engine that includes retries and backoff strategies. The significance of Cordum lies in its comprehensive approach to operationalizing AI agents, providing teams with tools that emphasize control and compliance. By integrating policy checks upfront and facilitating easy interaction with both API and CLI, Cordum aims to streamline the deployment and management of autonomous systems. With its focus on safety and auditability, it addresses a critical need in the AI/ML community for reliable and transparent automation. The platform also promises to evolve with enterprise features, enhancing its value for organizations seeking robust solutions for AI governance and operational control.