Overview / Description
Cordum is an AI agent governance control plane that enforces policy, manages human approval workflows, and maintains full audit trails for autonomous AI agents running in production. It is built for engineering and DevOps teams that need to govern agent actions before they execute — not after something goes wrong.
At the core of Cordum is a Safety Kernel that evaluates every agent job against policy-as-code rules before the job can run. Decisions are one of four outcomes: ALLOW, DENY, REQUIRE_APPROVAL, or ALLOW_WITH_CONSTRAINTS. Each decision is snapshot-based and replayable, and teams can simulate policy changes before rolling them out.
When an agent reaches a high-risk action, Cordum's human-in-the-loop approval gates pause execution and route the request to the right operator. Routing is risk-aware, decisions are bound to the job hash and policy snapshot for tamper-evident records, and the operator experience is designed for fast, clear responses.
Every action and decision is captured in a full audit timeline, so teams no longer need to stitch together logs from multiple tools when something needs review. Audit retention ranges from 7 days on the free Community plan to unlimited on Enterprise.
Cordum ships plugins for OpenAI, Anthropic, LangChain, LlamaIndex, CrewAI, AutoGen, AWS, Google Cloud, Azure, Temporal, Mistral, Cohere, Vertex AI, Bedrock, n8n, and Zapier, covering most production AI agent stacks. It can be deployed in minutes with no credit card required on the Community tier, and supports managed or on-premises deployment for enterprise customers.
Used For
AI agent policy enforcement in production, human-in-the-loop approval workflows for autonomous agents, audit trail generation for AI agent actions, risk-aware approval routing for high-stakes agent jobs, governing LangChain and CrewAI agent pipelines, compliance controls for AI-driven infrastructure automation, incident response governance for autonomous agents, SIEM integration for AI agent audit logs, on-premises deployment of AI governance controls, pre-execution policy simulation and testing
Pricing
Pros & Cons
Pros
- Policy-as-code Safety Kernel evaluates every agent job before execution with ALLOW, DENY, REQUIRE_APPROVAL, or ALLOW_WITH_CONSTRAINTS outcomes and snapshot-based replayable reasoning
- Human-in-the-loop approval gates with risk-aware routing, decision binding to job hash and policy snapshot, and multi-approver support on Team and Enterprise plans
- Integrations with 16+ major AI platforms and clouds including OpenAI, Anthropic, LangChain, CrewAI, AutoGen, AWS, Azure, and Google Cloud
- Full audit timeline capturing every agent action and decision, with retention from 7 days (Community) up to unlimited (Enterprise) and SIEM export on higher tiers
- Community edition is fully functional and production-ready at no cost, with up to 3 workers, 500 requests/second, and Slack/GitHub community support
Cons
- Team plan pricing is not published publicly — you must contact sales for a quote
- Community plan is limited to 3 workers and 3 concurrent jobs, which may not suit teams with high job throughput
- Managed cloud service is in early-access rollout to select partners only, not generally available
- BUSL-1.1 license restricts reselling or offering Cordum as a managed service without an Enterprise license
Alternatives
Zenguard AI, Portkey AI Gateway, LangSmith, Guardrails AI, TrustPath