Skip to main content
The Komodor Agentic Operation Platform (KAOP) is a control plane for AI agents that do real operations work. It gives your team one place to build agents, connect them to the systems they need, run them safely, and prove what they did. This page explains what KAOP is for and points you at the fastest route in.

​
Where this comes from

Komodor has spent years helping enterprise platform teams operate Kubernetes: making sense of what changed, why a workload broke, and what to do about it. That work produced a large body of operational knowledge — the checks, correlations, and runbooks reliability engineers actually follow. KAOP is the next step. Agents can now carry out that operational work, but running them in production raises a different class of problem: what an agent is allowed to touch, what it did, whether its conclusion can be trusted, and what it cost. KAOP is the platform layer that answers those questions, built from the experience of running operations for enterprise teams.

​
Who it is for

Platform engineers, SREs, and developers who want to put AI agents to work on production operations — and need that to be governed, auditable, and measurable rather than a collection of scripts and one-off bots.

​
What you get

A fleet you control

Build your own agents or deploy Komodor-built ones, each with its own identity, permissions, and history.

Governed access to real systems

Connect Datadog, AWS, GitHub, Kubernetes, Slack and more once — agents use them without ever handling raw credentials.

An evidence trail

Every run records its inputs, tool calls, reasoning, output, and cost, so a conclusion can be audited rather than trusted blindly.

Prebuilt outcomes

Modules package operational work you can switch on — incidents, alert quality, cost, drift — instead of building it.

​
The operating loop

Everything in these docs sits somewhere in this loop.
The agent at the centre of a conveyor line while Build, Run and Optimize stations travel past it, all inside a Govern boundary.

The operating loop: Build, Run and Optimize turn in sequence, all inside Govern.

  • Build — author an agent with the SDK or deploy one of ours, give it skills and knowledge, and connect what it may reach.
  • Run — work arrives from a schedule, a workflow step, a chat message, or a module, and every execution produces a run with its evidence.
  • Optimize — read fleet health, grade output quality, and attribute spend, then improve the agents accordingly.
  • Govern — identity, permissions, guardrails, approvals, and the audit log apply across all three.

​
Two ways to reach your systems

Worth knowing early, because everything about tooling reduces to it: an agent reaches an external system either through a built-in integration from the catalog, or through the MCP Gateway for your own tools. There is no third path.

​
Where to go next

Before you begin

The prerequisites — access, a connected system, and the SDK if you will author agents.

Concepts & glossary

The vocabulary the rest of the documentation is built on. Read it once.

Build your first agent

Create and run an agent, then read its evidence trail.

Architecture

How the control plane and your agents fit together.