Ungoverned autonomy
Agents can restart services, write to production systems, or push code without explicit policy approval.
Deterministic AI Control Plane
The Control Plane for Autonomous AI Agents
Ship autonomous AI agents safely. Enforce policy before execution, require human approvals where risk demands it, and keep a full audit trail from first action to final result.
Policy decisions
Every action evaluated before dispatch
Human approvals
Required for risky operations
Audit trail
Immutable timeline for compliance
Coordinate multi-step flows with deterministic execution state.
The Problem
AI agents are going autonomous. Governance can’t wait.
Teams are deploying autonomous AI agents fast. Without a control plane, risk and ambiguity scale faster than value.
No reliable audit trail
When someone asks what happened, teams stitch together logs across tools and still miss key decisions.
Shadow automation risk
Teams build one-off safety checks under pressure. Control is inconsistent and hard to review.
The Solution
The control plane for Autonomous AI Agents
Cordum gives you a single governance layer between agent intent and production action — enforce policy, require approval, and audit every decision.
Safety Kernel
Policy-as-code evaluates every job before it can execute.
- ALLOW, DENY, REQUIRE_APPROVAL, or ALLOW_WITH_CONSTRAINTS
- Snapshot-based decisions with replayable reasoning
- Simulation support before policy rollout
Human-in-the-loop
Approval gates pause high-risk operations until the right person approves.
- Risk-aware approval routing
- Decision binding to job hash and policy snapshot
- Fast, clear operator experience
Full audit timeline
Every action and decision is captured in a deterministic run history.
- Append-only execution records
- Searchable decision context
- SIEM-friendly export paths
Extensible packs
Add domain workflows and workers without destabilizing the core platform.
- Installable overlays for policy and config
- Scoped schema and workflow namespaces
- Disable or purge with clear rollback paths
How It Works
From job to result—governed every step
Policy checks, approval gates, and execution telemetry are built into the workflow lifecycle.
- 1
Job submitted
submitAn AI agent submits a job with context pointers and risk metadata.
- 2
Safety check
checkThe Safety Kernel evaluates policy in milliseconds before any dispatch happens.
- 3
Approval when needed
approveHigh-risk jobs pause until an authorized operator approves or rejects the action.
- 4
Worker execution
executeScheduler routes to capable workers with retries, timeout controls, and backpressure.
- 5
Audit sealed
auditRun results and decisions are written to immutable timeline records for review.
workflow.yaml
name: incident-remediation
steps:
collect_signals:
type: worker
topic: job.ops.collect
approval_gate:
type: approval
depends_on: [collect_signals]
restart_service:
type: worker
topic: job.ops.restart
depends_on: [approval_gate]
publish_audit:
type: notify
depends_on: [restart_service]Approval steps are first-class workflow nodes, not custom glue code.
Architecture
Production-ready architecture
Built for teams that need predictable behavior under pressure.
API Gateway
Unified HTTP, WebSocket, and gRPC control plane surface for jobs, runs, approvals, and policy.
Realtime stream support
Scheduler + Safety Kernel
Least-loaded routing with policy enforcement, budget checks, and stale-job reconciliation.
Deterministic dispatch
Workflow Engine
DAG execution model with retries, dependency handling, and run-level timeline tracking.
Parallel steps + failure semantics
NATS + Redis backbone
Message durability, pointer-based state, locks, and artifact metadata for production-grade agent governance.
JetStream-ready
Built on battle-tested foundations
NATS
Cloud-native event bus for fast, resilient workflow messaging.
Redis
High-performance state store for workflow pointers and metadata.
Go services
Predictable binaries for scheduler, kernel, workflow, and gateway.
Kubernetes ready
Deployable in modern platform stacks with clear operational boundaries.
GitHub
Live
Stars on cordum-io/cordum
View repository
NATS backbone
Resilient event streaming for jobs and workflow state changes.
Redis state
Fast pointer-based state and timeline metadata storage.
Go services
Predictable control-plane binaries built for production workloads.
Kubernetes ready
Fits modern platform stacks with clear operational boundaries.
Source-available. No black boxes.
Inspect the core code, review protocol details, and build with confidence.
- Full core source and CLI available on GitHub
- Published protocol specification and docs
- Community-friendly contribution path
Enterprise ready
Designed for organizations that need stronger control, support, and governance.
- SSO / SAML and advanced role controls
- Compliance-focused audit retention and exports
- Priority support and rollout partnership
Editions
Start free, then scale through Team and Enterprise.
Choose the path that matches your deployment stage and governance needs.
Community
For individual builders and internal teams validating autonomous AI agents.
- Full control plane (Safety Kernel, Scheduler, Workflow Engine)
- Dashboard & CLI
- Up to 3 workers
- Community support (Slack & GitHub)
Team
Coming Soon
Expanded capacity and collaborative governance for teams running multiple agents.
- Everything in Community
- Up to 25 workers
- Multi-approver gates
- 90-day audit retention
Enterprise
Recommended
SSO, compliance-grade audit controls, and SLA-backed support for governing AI agents at scale.
- SSO / SAML and advanced RBAC
- Unlimited audit retention + SIEM export
- Priority support with response SLAs
- Managed or on-prem deployment options
Newsletter signup
Frequently Asked Questions
What is Cordum used for?
Cordum is used to govern autonomous AI agents in production. It provides pre-execution policy checks, approval gates for risky work, and a full audit timeline of every decision and action.
How is Cordum different from an agent framework?
Agent frameworks focus on reasoning and chaining tasks. Cordum is the control plane that governs whether actions are allowed, who must approve them, and how they are audited.
Is Cordum open source?
Cordum is source-available under the Business Source License (BUSL-1.1). You can view, clone, run, and modify the core for internal use. Enterprise capabilities are available for organizations that need expanded governance and support.
What does a production setup need?
A standard deployment uses Go services with NATS and Redis. Production environments typically enable durable messaging, timeout reconciliation, and policy governance workflows.
Ready to ship autonomous AI agents safely?
Start in minutes with the quickstart, or talk with our team about enterprise governance needs.