Windows Agent (Installer)

This guide installs the NeuroStrike agent on a Windows host using the signed MSI installer. The same package works for a double-click install on one server and a silent, fleet-wide rollout through Intune, SCCM, or Group Policy.

Before you beginCopy Anchor Link

Choose a deployment profileCopy Anchor Link

The installer applies safe defaults based on the profile you pick. You can still adjust individual components.

Install (interactive)Copy Anchor Link

Enroll the agentCopy Anchor Link

The installer runs this as a step before it finishes, so the agent is enrolled the moment setup completes. You can also run it later with neurostrike-agent activate. How you authenticate depends on whether the host has internet.

Connected (relay)Copy Anchor Link

No token is ever typed into the installer — approval happens in your dashboard.

neurostrike-agent activate

Air-gapped (offline)Copy Anchor Link

The host never touches the internet, so it activates with a small licence file instead of signing in. Nothing is installed from the USB — the agent runs on the host; the USB just carries files in and out.

Silent / fleet deploymentCopy Anchor Link

For Intune, SCCM, or Group Policy, drive every choice through MSI properties — no UI is shown. Property values are case-sensitive: PROFILE is standard, airgapped, or ot.

msiexec /i NeuroStrikeAgent.msi /qn PROFILE=ot ADDLOCAL=CoreAgent RELAYURL=wss://agents.neurostrike.com

msiexec /i NeuroStrikeAgent.msi /qn PROFILE=standard ADDLOCAL=CoreAgent,Nuclei

VerifyCopy Anchor Link

neurostrike-agent doctor

A connected agent shows ONLINE in the dashboard. An air-gapped agent can run a first host assessment:

neurostrike-agent assess --require-roe=false
neurostrike-agent report --out assessment.html

Adding tools laterCopy Anchor Link

The agent is a thin runner over a wide tool catalog and fails soft when a tool is absent. Core host assessment needs no external tools; network discovery needs nmap, and template scans need nuclei. The broader offensive toolset (impacket, netexec, and similar) is bring-your-own and, being largely Linux-native, is best run from a Linux agent assessing Windows targets over the network.

Custom instructions & guardrailsCopy Anchor Link

You control what the agent may do and how it behaves through its engagement file and a plain-text instructions file — no coding, and most of it optional. The full list is in the Configuration reference; the essentials are below. There are two kinds of rule, and the difference matters:

• Hard limits (enforced). Anything the agent must never touch goes in scope.deny — those hosts are not probed, resolved, or even reported. Cap how intrusive it can get with limits.max_class (read-only → exploit), block specific tools with tools.deny, and set allowed hours. These are enforced in code.
• Guidance (advisory). Free-form direction — focus areas, tone, things to prefer or avoid — goes in guidance: or a plain .neurostrike/instructions.md file. The agent reads it as context.

scope:
  allow: [10.20.0.0/16]
  deny:  [10.20.9.0/24]            # never touched
limits:
  max_class: low_impact_active     # no intrusive/exploit tooling
  hours:
    timezone: Europe/Oslo
    windows:
      - { days: [Sat, Sun], from: "02:00", to: "06:00" }
tools:
  deny: [sqlmap]                    # block specific tools
guidance: >
  Focus on the public web tier and the VPN gateway. Prefer verification over
  breadth. Do not attempt credential brute-forcing.

# House rules
- Treat anything under /admin as sensitive — report, don't exploit.
- Our change window is 02:00–06:00 CET; avoid noisy scans outside it.

These files live beside the agent's data in C:\ProgramData\NeuroStrike and apply to every run on this host.

Human-in-the-loop approvalCopy Anchor Link

For hands-on or sensitive engagements, keep a person in control of every risky action:

• Require approval per action. On the on-prem AI relay, set REQUIRE_APPROVAL=true and APPROVAL_MIN_CLASS (default intrusive_active; read_only → exploit, or all). Each proposed command at or above that level pauses for an operator to approve on the console before it runs — exploit-class and AI-generated payloads always require approval.
• Record supervision. Set supervision: attended in the engagement file when a human oversees the run.

The installer's Oversight & rules step can turn these on for you; you can also change them later.

Older and legacy hostsCopy Anchor Link

The native agent needs Windows Server 2016 / Windows 10 or later. Older machines — Server 2012 R2, 2008, or Windows 7, still common on OT and legacy estates — are covered without installing anything on them:

• Run the agent on a nearby modern host (or a small Linux box) on the same network.
• Add the legacy machine as a target; the agent assesses it over the network — ports, services, SMB, and known vulnerabilities.

This keeps fragile or unsupported hosts untouched while still bringing them into scope. For deep on-host (assumed-breach) checks that need local access on a legacy OS, ask us about a legacy build.

Updating and supportCopy Anchor Link

Apply updates by installing the newer MSI; your state in C:\ProgramData\NeuroStrike — audit chain, engagement files, licence, and templates — is preserved. Security updates are provided for the committed support period.

UninstallCopy Anchor Link

Uninstall from Apps & features (Settings → Installed apps → NeuroStrike Agent → Uninstall) or with msiexec /x NeuroStrikeAgent.msi. A standard uninstall:

• stops and removes the Windows service,
• removes the program files, Start-menu shortcut, firewall rule, environment variables and registry keys,
• removes credential remnants — the enrollment token, the console operator password and the local-model override are always deleted, so an uninstalled host leaves no secret behind,
• retains the records — findings and the tamper-evident audit chain are kept (or exported) as a compliance record rather than silently deleted.

Destroying the data tooCopy Anchor Link

The records are kept by design. To remove them as well, choose one of:

• Certified destruction (recommended for engagement close-out). Before uninstalling, run the agent's purge command with its confirm flag. It destroys the findings and state and writes a dated destruction certificate into the audit chain — the written evidence a data-processing agreement's retention and destruction clauses ask for.

• Full wipe at uninstall. Pass REMOVEDATA=1:

  Copy Text

  msiexec /x NeuroStrikeAgent.msi REMOVEDATA=1 /qn

  This recursively deletes C:\ProgramData\NeuroStrike, records included. Use it only when a certificate is not required.

TroubleshootingCopy Anchor Link