cordum  by cordum-io

Governance-first control plane for autonomous AI agents

Created 3 weeks ago

New!

527 stars

Top 60.0% on SourcePulse

GitHubView on GitHub
Project Summary

Cordum addresses the "Trust Gap" between non-deterministic AI agents and production infrastructure by providing a governance-first control plane. It enables enterprises to safely integrate autonomous AI agents by enforcing Policy-as-Code, deterministic scheduling, and durable execution, filling a critical need for production-ready AI governance.

How It Works

Cordum inverts the traditional AI agent execution model. Instead of direct tool calls, it intercepts LLM intents as jobs, evaluates them against a Safety Kernel using defined YAML policies, and only dispatches approved commands to workers via a durable NATS JetStream bus. This infrastructure-based safety layer ensures that even compromised agents cannot execute forbidden actions, providing deterministic and crash-safe workflows orchestrated with Redis for state management and CAP v2 for wire contracts.

Quick Start & Requirements

Installation is streamlined via Docker Compose (docker compose up -d) or a one-liner script (curl -fsSL https://raw.githubusercontent.com/cordum-io/cordum/main/tools/scripts/install.sh | bash). For Kubernetes, Helm charts are available. Prerequisites include Docker and Docker Compose; Go is optional for cordumctl. curl and jq are needed for smoke tests. A 2-minute guardrails demo and a 1-minute quickstart are available, with detailed walkthroughs and installation guides in docs/demo-guardrails.md, docs/helm.md, docs/production.md, and docs/configuration.md.

Highlighted Details

  • Universal Orchestrator: Manages multi-step workflows (DAGs) with built-in retries, backoffs, and crash-safe state.
  • Intelligent Scheduler: Features least-loaded scheduling and capability-aware routing (e.g., GPU-specific jobs).
  • Pack System: Allows bundling of Workers, Workflows, and Policies into distributable overlays for modular deployment.
  • MCP Native: Supports the Model Context Protocol, enabling governance for existing tools via IDE agents.
  • Flight Recorder: Provides comprehensive logging for every decision and action, enabling replay of failed workflows.

Maintenance & Community

The README does not detail specific community channels (e.g., Discord, Slack) or notable contributors. Development appears active with separate repositories for enterprise features (cordum-enterprise) and official packs (cordum-packs).

Licensing & Compatibility

Cordum is licensed under the Business Source License 1.1 (BUSL-1.1). This license permits free self-hosted and internal use but is proprietary for competing hosted or managed offerings, restricting commercial service provision.

Limitations & Caveats

Advanced enterprise features such as SSO, RBAC, and SIEM integration are housed in the cordum-enterprise repository and necessitate a commercial license. The BUSL-1.1 license imposes restrictions on offering Cordum as a managed service.

Health Check
Last Commit

4 days ago

Responsiveness

Inactive

Pull Requests (30d)
39
Issues (30d)
56
Star History
554 stars in the last 26 days

Explore Similar Projects

Starred by Elie Bursztein Elie Bursztein(Cybersecurity Lead at Google DeepMind), Chip Huyen Chip Huyen(Author of "AI Engineering", "Designing Machine Learning Systems"), and
7 more.

SuperAGI by TransformerOptimus

0.1%
17k
Open-source framework for autonomous AI agent development
Created 2 years ago
Updated 1 year ago
Feedback? Help us improve.