• SOC 2 Readiness
  • 24/7 incident line
  • 24/7 security operations team
GuardsArm Platform

Agents & supported platforms

Enroll, configure and upgrade endpoints at scale, with an agent that defends itself.

The agent is both the sensor the detection engine runs on and the actuator response is executed through. It is distributed via signed package repositories and a hosted Windows installer.

  • Linux · Windows · ARM64
  • Signed packages
  • Self-updating
Lightweight endpoint agents running on Linux, Windows and ARM64 systems, reporting telemetry upward and receiving response actions back

GuardsArm Agents

The one part of the platform that lives on your hosts

Everything else in GuardsArm runs on servers you set aside for it. The agent is different: it sits on the machines you are actually trying to protect, which is why it gets more scrutiny from IT and change boards than any other component. This page answers the questions that scrutiny tends to raise: where it runs, how it gets onto a fleet, what it reads, how it is kept current, and what stops it being switched off.

In short, GuardsArm Agents collect host activity and send it to your own GuardsArm Manager, where detection and correlation happen. The same agent carries response back down: isolating a host, killing a process, quarantining a file. One install gives you both visibility and the means to act on it, and none of that traffic leaves infrastructure you control.

Platform support

Where the agent runs

PlatformStatusNotes
Linux, Debian / Ubuntu (.deb)ShippedFull EDR, file integrity monitoring, configuration assessment, inventory and response.
Linux, RHEL / Rocky (.rpm)ShippedFull parity with the Debian package.
Windows, signed .msiShippedCode-signed, with kernel-level telemetry, on-agent tamper protection and network isolation for host containment.
ARM64, Linux .deb and aarch64 .rpmShippedFor ARM servers and edge deployments.
macOS (.pkg)RoadmapCode is ready; pending a signed and notarized build. Not yet generally available.

macOS is not yet generally available. The code is ready and the build is pending signing and notarization - it is listed here so the gap is explicit rather than discovered during a deployment.

How it fits the platform

Mixed estates are the norm. The Linux packages have full parity across Debian-family and Red Hat-family distributions, and ARM64 builds cover ARM servers and edge hardware, so one policy model spans the data centre, the cloud fleet and the branch.

Deployment and enrolment

How a rollout actually runs

Five steps, in the order that keeps a rollout boring. Most of the effort is in the first one.

  1. Plan the groups first

    Decide how the estate divides before anything is installed: by role (domain controllers, web servers, workstations), by environment, or by customer if you run GuardsArm for several organisations. Groups carry configuration, so this decision shapes everything that follows.

  2. Issue an expiring enrolment token

    Enrolment uses a token with a limited life rather than a permanent shared secret. A token that leaks after the rollout window has closed is useless to whoever finds it.

  3. Install through the channel you already use

    Linux hosts install from signed package repositories, so the agent arrives the same way as any other package your configuration management already handles. Windows hosts take a code-signed installer that fits standard software-distribution tooling and golden images.

  4. Enrol and inherit configuration

    On first contact the agent registers with your GuardsArm Manager, joins its group and picks up that group’s collection and response settings. Nothing is configured host by host.

  5. Confirm health before widening

    The fleet health view shows which agents are reporting and which have gone quiet. Start with a pilot group, confirm every host checks in, review what it sends, then extend group by group.

What it collects

Three questions the agent answers about every host

Described by what it tells you rather than how it gathers it. What each group collects is set on the group, so sensitive hosts can watch more and quiet ones less.

What is running

  • Process starts and the process that launched them
  • Outbound and listening network activity per process
  • New persistence: services, scheduled tasks, start-up entries

What changed

  • Real-time changes to files and paths you mark as sensitive
  • Drift from your configuration and hardening baselines
  • Suspicious files matched on the host, with quarantine

What the host is

  • Installed software and versions
  • Open ports and running services
  • Operating system, build and hardware facts

The inventory side matters more than it first appears. Because the agent keeps a continuously updated record of installed software, the vulnerability management module works from the same asset record the detections use, so a vulnerable package and a suspicious process on the same server resolve to one machine rather than two disconnected tickets. The same data underpins file integrity monitoring and asset management, which is also where auditors tend to ask for evidence.

Fleet management

Running the agent estate day to day

Configuration by group

Collection settings, monitored paths and response permissions live on the group. Changing what two hundred servers watch is one edit, and a new host is correct from the moment it enrols.

Remote agent upgrade

Agent versions are upgraded from the console, so moving the estate to a new release does not mean a reinstall pass or a ticket per machine.

Fleet health and statistics

Live health views and per-agent statistics make a silent agent visible. A host that has stopped reporting is a finding in its own right, not an absence nobody notices.

Signed content updates

Agents pull signed content updates on their own without a full reinstall, so detection content stays current between agent releases.

Customer and tenant mapping

For providers running several organisations, each agent maps to a customer group and its data is visible only within that tenant.

Clean, deliberate removal

When an agent is meant to come off a host, uninstall is clean and fails open, so decommissioning a server never leaves it broken or half-protected.

Tamper resistance and updates

Built on the assumption that someone will try to switch it off

Disabling endpoint security is one of the first things an intruder with administrative access tries. The agent is built with that in mind. Self-healing guardians watch for attempts to stop or disable it and reverse them, and its state, including any containment in force, persists across a reboot. Removal is still possible when it is genuinely intended; it is simply not something that happens quietly.

Updates follow two separate tracks. Content, the material the agent uses to recognise threats, is signed and pulled by the agents automatically, with no reinstall. Agent versions are upgraded remotely from the console, which means you decide when a new release reaches each group and can stage it on a pilot first. Packages, the Windows installer and content updates are all signed, so you can verify where every piece of agent software came from.

A sensor that can be silenced without anyone noticing is a log source, not a defence.

Before you roll out

Mistakes that slow agent deployments down

Enrolling everything into one group

It works on day one and becomes painful by month three, when servers and laptops need different monitored paths and different response permissions. Split by role from the start.

Switching on automatic response before baselining

Keep high-impact actions behind approval gates while a pilot group settles in. Knowing what normal administration looks like on your hosts is what keeps automated isolation pointed at attackers rather than your own scripts.

Treating a quiet agent as a healthy one

Check the fleet health view as part of routine operations. An agent that stops reporting after a rebuild or image change is the most common coverage gap, and the easiest to close.

Assuming every device needs an agent

Firewalls, switches, cloud services and SaaS platforms are better covered through their logs. Put agents where host-level visibility and response matter and send the rest to the platform as log sources.

Questions

Common questions about the agent

Which operating systems does the agent support today?

Linux on Debian and Ubuntu (.deb) and on RHEL and Rocky (.rpm), Windows through a code-signed .msi, and ARM64 Linux through .deb and aarch64 .rpm packages. A macOS agent is on the roadmap; it is not generally available yet, so plan Mac coverage separately for now.

Where does agent data go?

To your own GuardsArm Manager and nowhere else. The whole platform runs on infrastructure you control, so endpoint telemetry stays inside your perimeter. The platform does not depend on internet access to operate, which also makes it workable in restricted or air-gapped networks.

How heavy is the agent?

It is designed to be lightweight, and much of the work, such as correlation and incident building, happens on the platform rather than on the host. We do not publish fixed resource figures because they depend on how much each group is configured to collect. We size it against your real host mix during a walkthrough or pilot.

Can an attacker or a local administrator simply turn it off?

The agent includes tamper protection: self-healing guardians that detect attempts to disable it and reverse them. Its state and any response actions also persist across reboots, so restarting the machine is not a way to undo a containment.

How are agents kept up to date?

Two ways. Content updates are signed and pulled by the agents themselves, without a reinstall. Agent versions are upgraded remotely from the console, group by group, so you can stage a release on a pilot group before rolling it wider.

Do we need an agent on every device?

No. The agent is what gives you host-level visibility and on-host response, so it belongs on servers and workstations you care about. Network gear, cloud services and SaaS applications are covered by sending their logs to the platform instead. Devices that cannot take an agent are a log-source question, not a gap in the agent.

Roll it out across your estate

We will walk through enrollment, grouping and upgrade for the mix of platforms you actually run.