Sage Infrastructure Solutions Group IncSage Infrastructure Solutions Group IncLoading…

Sage Infrastructure Solutions Group Inc

Windows hosts

Keyreeve manages Windows hosts as well as Linux ones. On Windows the agent runs as a service and provides the terminal itself, so the host needs no OpenSSH. People reach it the same ways as a Linux host — the desktop app, the web terminal, or ssh through the Keyreeve gateway — and every session is recorded.

What a host needs

  • Windows Server 2019 or later, or Windows 10 1809 or later (Windows 11 included), on amd64 or arm64
  • Outbound access to the Keyreeve agent service address (port 4388 unless it is behind a TCP passthrough), over mutual TLS. The agent never accepts inbound connections.
  • An administrator to run the installer

Install

Create an enrollment token under Admin → Hosts → Enrollment tokens, then use Copy Windows install command and paste it into Windows PowerShell run as administrator on the host. The command downloads install.ps1 from your agent service and checks it against the SHA-256 written into the command. The script then:

  1. picks the build for the machine (amd64 or arm64) and downloads its MSI;
  2. checks the MSI's SHA-256 and its Authenticode signature against Sage ISG's release certificate;
  3. installs it to C:\Program Files\Keyreeve, enrolls the host and starts the keyreeved service.

It's safe to run again. As on Linux, a new host waits for approval unless its token approves hosts automatically.

To install the MSI yourself (from /download/keyreeved-windows-amd64.msi or -arm64.msi on the agent service):

msiexec /i keyreeved-windows-amd64.msi /qn SERVER=https://agents.keyreeve.example.com:4388 TOKEN=kret_… CAPIN=sha256:…

The agent updates itself like the Linux agent. Uninstalling (msiexec /x) removes the service and keeps %ProgramData%\Keyreeve unless you add PURGE=1.

Shells

People choose Windows PowerShell (the default), PowerShell 7 or Command Prompt: in the desktop app per host (the ✎ editor), on the web terminal's page, or with ssh -o SetEnv=KEYREEVE_SHELL=pwsh through the gateway. Nothing else is run.

Which account people log in as

Set under Admin → Fleet settings → Windows accounts (host groups and hosts can override it):

SettingMeaning
Where accounts come fromAutomatic (the default): directory accounts on domain-joined hosts, Keyreeve-managed local accounts on standalone ones. Or always directory, or always managed
Naming templateHow a person's directory account is named on the host: {domain}\{sam} (the default), {sam}, {upn}
DomainThe {domain} part. Empty: the domain the host is joined to

Directory accounts (domain-joined hosts)

People log in as their own Active Directory account. Keyreeve takes the account name from their sign-in token: a claim named samaccountname, onprem_sam_account_name or onPremisesSamAccountName (or windowsaccountname as DOMAIN\name), and the UPN from onPremisesUserPrincipalName, onprem_upn or upn. Configure your identity provider (Entra ID claims mapping, a Keycloak or ADFS mapper) to send one. Without any of them, Keyreeve uses the part of their identity before the "@". Keyreeve never creates, disables or deletes directory accounts. If the account doesn't exist on the host, the login is refused with that reason.

Keyreeve-managed local accounts (standalone hosts)

With account creation turned on in the fleet settings, the agent creates a local user per person, named after their identity (at most 20 characters; a -2 suffix if the name is taken), with a long random password nobody knows. By default the account may not sign in at the console or by Remote Desktop: the only way in is through Keyreeve. The account is disabled when access ends and enabled again if it returns, and is deleted with its profile after Delete idle accounts after (days) if you set it. Keyreeve only ever touches accounts it created, and never creates local accounts on a domain controller.

An account a rule names explicitly (for example Administrator) is used as named, in either mode.

Administrator rights

Windows has no sudo. In an access rule (step 4, Privileges), or a temporary-access policy, you can grant on Windows hosts:

  • Elevated sessions only — that person's Keyreeve sessions run with administrator rights; their account never joins any group;
  • Administrators group — the account is in the local Administrators group while access lasts, and Keyreeve takes it out again when it ends (only memberships it added).

Both have to be confirmed when you save. On a domain controller, Administrators membership is refused (the group there is the domain's); elevated sessions still work and give that session full control of the domain, which the confirmation says.

Recording

Every byte passes through the agent, so every session is recorded with the same protections as on Linux (Session recording). Each session's processes run in a Windows job object, so ending the session ends everything it started, including when temporary access runs out. The recording shows the Windows account used and whether the session was elevated. The agent also logs the programs each session ran, from Windows' process-creation events (it turns that auditing on itself and puts the policy back as it was if you turn it off).

Someone at the keyboard

Admin → Fleet settings → Windows: someone at the keyboard (per host group or host too) decides what the person signed in at the machine sees when someone opens a Keyreeve session: automatic (a notice on workstations, nothing on servers), nothing, a notice, or ask — they must accept within the timeout (10–600 seconds, 60 by default), and no answer refuses the session.

Other ways in

Windows hosts always report console, Remote Desktop and network logons that didn't come through Keyreeve, in Audit → Logins. Under Admin → Fleet settings → Windows: other ways in you can also allow console and Remote Desktop logons to the host's break-glass accounts alone. Network logons (file shares, WinRM) aren't blocked, because Keyreeve's own logon is one. It isn't applied on domain controllers or on a host with none of the break-glass accounts, and in a domain, Group Policy that sets the same rights will fight it.

Break-glass

Windows break-glass policies (Admin → Break-glass, Windows hosts) name existing local accounts allowed in at the console or by Remote Desktop when Keyreeve can't be reached, with their password either:

  • the host's own — Keyreeve never touches it; or
  • a Keyreeve-managed rotating password — LAPS-style: a long random password, sealed to the Keyreeve servers, rotated after each use. An administrator can reveal it once under Windows break-glass passwords, and every reveal is audited.

Every break-glass logon is reported and can notify the email addresses and webhook on its policy, as on Linux (Break-glass access).

Limitations

  • No network credentials in a session. Keyreeve logs people on without a password, so a session can't reach file shares, SQL Server with Windows sign-in or other servers ("double hop").
  • Legacy scp (which runs scp on the host) doesn't work; use sftp, or a current scp, which uses SFTP.
  • Logging on as an Entra ID (cloud) account on Entra-joined machines is not yet verified.