A host-based intrusion detection system (HIDS) is software that runs on the server itself and watches for signs that someone has broken in or is trying to. Where a network intrusion detection system watches traffic passing a point on the network, a HIDS sees what actually happens on the machine: who logged in, what ran, what connected out and which files changed.

What a HIDS watches

  • Logins: successful and failed SSH logins, new users, sudo use, logins from places or at times that don’t fit.
  • Processes: new or unusual programs, especially ones started by a web server or running from /tmp.
  • Network connections: listening ports that appear, and outbound connections to addresses that have no business being contacted.
  • File integrity: changes to system binaries, SSH keys, cron jobs, systemd units and other places attackers use to stay in.
  • Logs: authentication, kernel and application logs, matched against known attack patterns.

Most real intrusions on Linux servers leave traces in several of these at once: a brute-forced login, then a download, then a new cron entry or SSH key so the attacker can come back. A HIDS is useful because it can connect those steps on the host where they happen.

HIDS vs NIDS

Host-based (HIDS) Network-based (NIDS)
Runs on Each server A network tap, span port or gateway
Sees Logins, processes, files, local logs Packets and flows between machines
Encrypted traffic Sees what happens after it is decrypted on the host Mostly sees metadata only
Blind spot Hosts without an agent What happens inside a host
Fits Cloud and VPS servers you don’t control the network for Networks you own end to end

For servers rented from a cloud or VPS provider you usually can’t put a sensor on the network at all, so host-based detection is the practical option.

HIDS vs EDR vs SIEM

  • HIDS: detects and alerts on activity on one host. Classic examples are OSSEC and file integrity checkers like AIDE.
  • EDR (endpoint detection and response): a HIDS plus the ability to respond, such as killing a process or isolating a machine, usually with a central console. Many EDR products were built for laptops and Windows first.
  • SIEM: collects logs from everything into one place to search and correlate. It depends on agents or log shippers on the hosts to feed it, and someone has to read what it finds.

Options for Linux servers

Tool What it is Good to know
OSSEC The original open-source HIDS: log analysis, file integrity, rootkit checks Mature and still actively released
Wazuh Open-source platform that grew out of OSSEC, with agents, a manager, an indexer and a dashboard Very capable; you run and maintain the server side
AIDE File integrity checker that compares files against a stored database Detects changes only; needs scheduling and someone to read reports
auditd The Linux kernel audit framework Records almost anything; turning records into alerts is up to you
Falco Runtime detection from system calls, aimed at containers and Kubernetes Rules engine; alerts need routing somewhere
osquery Exposes system state as SQL tables Great for asking questions; not an alerting product on its own
Vigilhold Hosted agent and console from Al’s-Ops Detection plus limited response you approve; from $39/month for five servers

Tools like fail2ban, SSHGuard and CrowdSec are often mentioned in the same breath. They block addresses that attack a service, which is prevention for one kind of attack rather than intrusion detection, but they pair well with a HIDS. Compare fail2ban alternatives

What to look for

  1. Alerts a person will read: grouped repeats and plain explanations, not one message per log line.
  2. Coverage you can check: a way to see that every server is reporting, and a warning when one goes quiet.
  3. Response you control: automatic actions limited to safe ones, with anything disruptive waiting for approval.
  4. Setup time that matches your team: an open-source stack is free to download but someone has to run it.

Where Vigilhold fits

Vigilhold is a host agent for Linux servers, including Docker hosts, with one web console for all of them. It watches logins, processes, connections and file changes, explains critical and warning alerts in two sentences, and asks before anything disruptive. When a response playbook bans an address, the ban reaches every server in the account. It installs with one command per host. How Vigilhold works

A HIDS answers one question a firewall can’t: what actually happened on this server?

Originally published on the Al's-Ops blog.