Cordum — An action firewall for AI agents, enforcing policy and human approval before risky operations.
Analyzed by Sai Pavan Gopularam · AI · Governance · View on GitHub
- Stars: 509
- Forks: 34
- Commits last 30 days: 28
- Health: Active (28 commits this month)
- Language: Go
- License: Other
What It Is
Imagine a bouncer standing in front of your AI agents, checking every action they try to take before they actually do it. That's Cordum. It's an 'Agent Control Plane' that acts like a firewall for your AI, letting you set rules for what your agents can and cannot do.
As AI agents gain more autonomy and access to real-world systems (like cloud accounts or databases), they can accidentally or intentionally cause significant damage. Cordum solves this by providing a critical layer of oversight, preventing unwanted actions and requiring human approval for sensitive tasks, ensuring safety and compliance.
License Verdict
BUSL 1.1 License — Limited Commercial Use — Commercial Use Restricted for Competitive Products • Source Available
The Business Source License 1.1 allows you to use, modify, and distribute the code for non-production purposes. For production use, it's generally permitted unless your product is directly competitive with Cordum's offerings, which could involve providing agent governance as a service. Review the full license terms carefully for specific restrictions.
How to Use It
Cordum offers two main quickstart options: a one-command script to install from source, or using Docker Compose with pre-built images. Both set up the full stack including the dashboard, API gateway, and various engines.
Prerequisites:
- Docker Desktop v4+ (or Engine v20.10+)
- Go 1.26.3+
- curl
- jq (recommended for Option B)
Estimated setup time: 10 minutes.
git clone https://github.com/cordum-io/cordum.git
cd cordum
./tools/scripts/quickstart.sh
What I'd Build With This
AI Agent Compliance Policy Packs (micro-saas)
Develop and sell pre-configured Cordum policy packs for specific industry compliance needs (e.g., HIPAA for healthcare agents, SOC2 for financial agents). Founders would pay for curated, ready-to-deploy YAML policies that integrate with their Cordum instance, saving them policy engineering time. Market via developer forums and compliance blogs.
Effort: 1 Week Build Time · Target: DevOps teams, Compliance Officers · Pricing: $199/pack one-time
Hosted AI Agent Governance Platform (saas)
Offer a managed Cordum service where businesses can connect their AI agents for centralized policy enforcement, human-in-the-loop approvals, and audit trails. This removes the operational burden of self-hosting. Charge based on agent activity (e.g., number of actions processed, approvals requested). Target companies adopting AI agents but lacking internal governance expertise.
Effort: 3 Months Build Time · Target: Mid-market Enterprises, AI-first Startups · Pricing: Starts at $499/mo
Custom AI Agent Control Plane & Integration (enterprise)
Provide consulting and custom deployment services for large enterprises. This involves integrating Cordum into existing complex infrastructure, developing bespoke policy engines for unique business rules, and offering specialized training and support. Focus on industries with high regulatory requirements or significant risk from autonomous operations, like finance or critical infrastructure.
Effort: 6 Months+ Engagement · Target: Fortune 500, Government Agencies · Pricing: $50,000+ per project
Sai Pavan Gopularam's Take
Cordum is a smart piece of infrastructure for anyone building with AI agents. The 'action firewall' concept is exactly what's needed to prevent AI agents from going rogue. If I were building a business around this, I'd focus on selling pre-built compliance policy packs for specific industries, easily generating $5-10k/month within a year.
Watch Out For
- BUSL 1.1 License Restrictions: The Business Source License (BUSL 1.1) limits commercial use, especially if your product competes directly with Cordum. This means a direct SaaS offering of Cordum might be problematic without a commercial agreement.
- Resource Demands: Running the full Cordum stack, especially with Docker, requires at least 4 GB of RAM allocated to Docker Desktop. Under-resourced systems might experience performance issues or instability.
- Policy Complexity: Defining effective and comprehensive YAML policies for AI agent governance can be complex. It requires a deep understanding of potential agent actions and security implications to prevent unintended consequences.
- Default Credentials: The quickstart uses default `admin` / `ChangeMe123!` credentials. These must be changed immediately for any non-local or production deployment to prevent security vulnerabilities.
I break down trending repos like Cordum every week — join the newsletter.