Cordum
Cordum is an out-of-process governance control plane designed for autonomous AI agents, providing a robust layer of policy enforcement, approval workflows, and cryptographic audit trails. It acts as a sidecar or proxy that intercepts agent actions before they are executed, evaluating them against a set of configurable policies. This pre-dispatch enforcement ensures that agents cannot perform unauthorized operations, such as accessing sensitive data, making changes to infrastructure, or executing high-cost transactions without appropriate checks. Built for production environments, Cordum enables organizations to define granular governance rules using a policy engine, and integrates human-in-the-loop approval gates for critical actions. It generates HMAC-signed audit logs that provide a tamper-evident record of all agent decisions, facilitating compliance with regulatory requirements and internal auditing standards. The system includes a dashboard for real-time visibility into agent governance decisions, allowing operators to monitor policy violations and approval histories. Cordum's out-of-process architecture means it can be deployed independently of the agent runtime, supporting polyglot agent implementations without requiring code modifications. It integrates via SDKs or service meshes, ensuring minimal performance impact while maintaining a strong security boundary. The tool is open source and actively developed, with features such as a safety kernel that defaults to fail-closed operation, injection scanners, and per-agent decision views.
- Category
- AI Agents
- Pricing
- Open Source
Problem solved
Organizations deploying autonomous AI agents face significant risks: agents may execute unintended or harmful actions, violate compliance policies, or make irreversible mistakes without oversight. Traditional logging and manual reviews do not prevent these failures in real-time. Cordum solves this by providing a pre-dispatch governance layer that enforces rules, requires approvals for sensitive actions, and securely records all agent decisions, enabling safe and auditable agent operations in production environments.
Use cases
- Enforcing data privacy policies to prevent a customer-facing agent from exposing personally identifiable information (PII).
- Requiring human approval before an agent modifies cloud infrastructure or makes financial transactions.
- Auditing all actions of an internal agent for regulatory compliance (e.g., HIPAA, GDPR).
- Implementing time-bound or scope-bound policies, such as limiting an agent's access to only certain APIs during business hours.
- Staging and testing governance rules before deploying agents to production, using the dashboard to simulate policy enforcement.
Audience
- Developers
- ML Engineers
- Data Scientists
- Enterprise
Pros
- Out-of-process integration without modifying agent code
- Pre-dispatch policy enforcement prevents unauthorized actions
- Cryptographic audit trails ensure tamper-proof logging
- Open-source with active community development
- Dashboard provides real-time governance visibility
Cons
- Requires integration effort to set up and configure
- May introduce latency due to policy evaluation
- Relatively new project with limited community size
Official website: https://github.com/cordum-io/cordum
Canonical LambdaBase page: https://www.lambdabase.com/tools/cordum