Why Wazuh

Something has to watch the machines themselves. Not the edge, where attackers arrive from the internet, but the hosts: files changing that should not change, packages with known vulnerabilities, logins that succeeded and should not have, a process doing something out of character.

Drafted by an AI agent (claude-opus-5) from this lab’s own runbooks, deployment log and errata. Reviewed before publication by the site owner. How this site is written →

That job goes to Wazuh, running on its own machine with an agent on every host in the fleet.

What it replaced

CrowdSec everywhere was the first design, and this project wrote about deleting it. It has no file integrity monitoring, no vulnerability scanning, and its crowd-sourced reputation means nothing on an internal network. It stayed at the edge and something else was needed inside.

Security Onion was the serious alternative: a full network security monitoring platform bundling signature-based intrusion detection, protocol analysis, packet capture, its own analyst interface, and Wazuh itself as the host-monitoring component.

It was evaluated properly and rejected, and the reason is the interesting part.

Why this one

Security Onion is built around seeing network traffic, and this architecture has nowhere to show it any.

Its core components watch packets crossing a chokepoint, which in a conventional network means a mirror port on a switch or a physical tap. Without that, the parts that make it what it is see nothing meaningful, and what remains is the host-monitoring component that Wazuh provides on its own.

This lab has no such chokepoint, and the reason traces directly to another choice on this blog. Public traffic does not arrive as routed packets crossing an internal switch. It arrives already terminated at a rented gateway and comes back through an outbound tunnel, emerging inside the network as a connection from a machine that is already on it. There is no point in the topology where a tap would see the traffic Security Onion exists to inspect.

That is worth stating carefully: not a worse tool, a tool whose central assumption this network does not satisfy. On a conventional office LAN with a core switch and a routed perimeter, the evaluation would have gone the other way.

The resource floor settled what was left. Security Onion’s own documentation puts its evaluation mode, explicitly labeled as not for production, at four processors and eight gigabytes, with a real deployment substantially above that. In a lab where the entire hypervisor is a shared budget, a security tool that consumes a large fraction of it to deliver its weakest half is a poor trade.

How it is actually used here

Live on a dedicated machine, with an agent on every host in the fleet, and the agent count verified from the manager rather than from the agents.

That detail matters more than it reads. An agent that believes it is connected is not evidence of anything: this project has an errata entry for exactly that, where starting the agent service was taken as proof a host was monitored. The check that counts is the manager’s own list of who is talking to it, and that is what the acceptance suite asserts.

What it actually watches, and the honest gap

Four capabilities carry most of the value here.

File integrity monitoring, which notices when something under the system configuration directories changes. On a fleet where every host is supposed to be described by configuration management, an unexpected change is either drift or something worse, and both are worth knowing about.

Vulnerability detection, which compares installed package versions against published advisories. This is the part CrowdSec cannot do at all.

Log analysis and correlation across hosts, which is where a pattern like the same failed login walking through several machines becomes visible.

Security configuration assessment, checking hosts against hardening baselines rather than trusting that they were built correctly once.

The honest gap: the manager’s own configuration and its agent registrations have no backup at the time of writing. Every agent enrolment lives on that one host, and the storage arrangement that would carry it offsite is blocked on hardware. A monitoring system that would have to be rebuilt from scratch after losing its own machine is a real dependency, and it is listed as such rather than quietly assumed away.

What it cost

It is the heaviest single service in the lab. A manager, an indexer and a dashboard, on a machine sized for them. That cost bought the four capabilities above, and it is the largest resource commitment made to anything that produces no user-facing function at all.

Its own monitoring is thinner than what it monitors. The check that watches it confirms that the path into it responds. That would stay green with the manager dead, the indexer refusing every login, and zero agents connected, and the record says so plainly rather than counting it as coverage. Closing it properly needs an internal monitoring path that does not exist yet, and the alternative, exposing an unauthenticated route into a dashboard with a serious vulnerability history, was rejected as worse than the gap.

Deploying it produced two errata before it worked. One where a value that looks like a per-host setting is only read when the package is first installed, so changing it never reaches an already-installed agent. One where a credential file at the obviously-correct restrictive permission was unreadable by the service that needed it. Neither is exotic. Both cost time.

What would change the decision

The agent’s reasoning from the record rather than the human’s stated position.

  • If the network gained a real chokepoint. The rejection of the fuller platform was about topology, not quality. A redesign that routed internal traffic through a point where a tap could see it would make that evaluation worth redoing.
  • If the resource cost stopped being justified by use. A security tool nobody reads is an expensive way to feel careful. The measure is whether its findings change anything, and that question deserves asking again once the fleet stops growing.

About Wazuh

Wazuh is an open source security platform combining host-based intrusion detection, file integrity monitoring, vulnerability detection, log analysis and security configuration assessment. It runs as a central manager with lightweight agents installed on monitored hosts, and includes its own indexing and dashboard components. It began as a fork of OSSEC and is released under the GNU General Public License v2.