Taking Agentic Operations
To the Next Level

Yep, Komodor is fully SOC2 compliant.

Unlike a passive monitoring agent, a Komodor Agentic Operations agent runs as a worker inside your own environment and reads what a given task needs — logs, metrics, configs, or the systems an integration exposes — to investigate and act. Any secret or credential value is automatically masked before it ever leaves the worker, so it can’t end up in a run record or log line. You control the scope: per-agent permissions decide what each agent can read or write and where, and you can further restrict access with your own RBAC and integration settings. We also employ agent guardrails to avoid prompt injections, unsafe tool calls, and secret and PII data crossing. 

After a lot of research, we found this to be the best solution for the majority of our users.

Yes. Agents run wherever your production runs — Komodor Cloud, your own Kubernetes, cloud on-prem, or air-gapped clusters — with the same platform and governance in every case. For environments with strict egress or data-residency requirements, the control plane itself can also run in your own VPC or on-prem, so nothing about the connection between your agents and the control plane leaves your network.

Komodor’s agent doesn’t transfer any PII or sensitive data, and allows full control of which data is uploaded.

A2A (“agent-to-agent”) is how one agent invokes another — for example, a workflow step calling a specialized agent to do part of the work. Gating means those calls run through the same control plane as everything else, so an agent can’t quietly invoke another agent outside your permissions and audit trail — every A2A call is authorized and logged like any other governed action.

Every action an agent proposes — a tool call, a write, an output — passes through a policy gate before it happens. Depending on your policy, that gate can allow it, redact part of it, block it outright, or hold it for a human to approve — which is how a destructive action (say, deleting a resource or force-pushing a change) gets caught before it executes rather than after. The same gate is what lets you catch sensitive content on the way out, not just gate what an agent can do on the way in.