Remote inventory

GLPI Agent supports to run computer inventory remotely, this feature can also be referenced as Agent-less inventory for targeted computers.


This task can inventory remote computers via:

  • ssh: for unix/linux computers

  • WinRM: for win32 computers


To remotely inventory unix/linux computers supporting ssh, GLPI Agent needs to make network requests on ssh port. The remote ssh user must have administration privileges.

For windows computers, WinRM must be enabled on targeted computers. You can follow Microsoft official documentation to enable WinRM. But the short way to enable it with minimal security is to run from an administor console:

winrm quickconfig
winrm set winrm/config/service/auth @{Basic="true"}
winrm set winrm/config/service @{AllowUnencrypted="true"}

You may probably need to also create a dedicated user with administrative privileges and enable the windows remote management firewall rule.

For WinRM, GLPI Agent must be able to make network http requests on WinRM dedicated ports, by default 5985 for HTTP and 5986 for HTTPS. The remote WinRM user must have administration privileges.


WinRM remote inventory can be run from an agent running on unix/linux platform.


Remote computers will be known as remotes on GLPI agent side.

Remote registration

So before running inventory, you'll have to register in GLPI Agent environment remote computers with dedicated credentials. This step can be done by glpi-remote script using the add sub-command.

This is as simple as running:

glpi-remote add ssh://admin:pass@


glpi-remote add winrm://admin:pass@ --target server0


When add a remote supporting WinRM, the agent will test provided credential and will fail if something goes wrong. You can avoid this check using -C or --no-check option.

Managing remotes

After remotes has been registered, you can list them with the following command:

glpi-remote list

This will provides the locally known remotes:

index  deviceid                 url                                      target   Next run date
    1  WIN-2020-09-23-15-37-52  winrm://glpi-agent:****@  server0  Tue Nov  9 15:46:51 2021
    2  XPS-2021-11-10-15-10-16  winrm://glpi-agent:****@  server0  on next agent run

You can delete a remote giving its listing index:

glpi-remote delete 1

You can update credential by simply register again the remote as the script will recognize your are updating an existing remote.


As of this writing, no solution has still been implemented in GLPI to manage remotes. So everything has to be done from the console.


Automatic execution

When run as a service or a daemon and once remotes are registered against GLPI Agent and associated to a target, the agent will run RemoteInventory task when expected, generate an inventory and submit it to the related server or store it to a local path.

The selected target must be a known target:

  • if selected target is server0, server must be defined in configuration,

  • if selected target is server1, server must be defined with at least 2 URLs as server1 means to use the second URL,

  • if selected target is local0, local must be set in configuration with an existing path.

Command-line execution

When GLPI Agent is run from the commandline, it will try to run RemoteInventory task if at least one remote is known. It will then select one remote and only one to run an inventory but only if its Next run date has been set to on next agent run.

You can try to run only RemoteInventory task with:

glpi-agent --logger=stderr --tasks remoteinventory

You may have to run again the agent if another remote is expected to be inventoried. Just run glpi-remote list to verify if a remote has to be inventoried.

Without remote registration

You also can use the --remote option of glpi-agent to process a remote without registering it:

glpi-agent --remote=ssh://admin:pass@ --logger=stderr --tasks remoteinventory

--remote option can be handy to schedule a remote inventory via crontab or windows job scheduling.


--remote value can be a list of remote url separated by commas. So commas are prohibited in passwords.

Performance with X remotes defined

By default, only one remote inventory can be run at a time.

Since GLPI-Agent 1.5, you can use --remote-workers option to set the maximum number of worker the remote inventory task can run at the same time, each worker processing one remote.

For example, the following command will process 2 remote inventory at the same time:

glpi-agent --remote-workers=2 --remote=ssh://,ssh:// --logger=stderr --tasks remoteinventory


In some context, you may need to change the way remote inventory is processed. In that case, you can configure your remote to use modes.

Modes must be set with the remote url itself to only be applied on one remote. The syntax is similar to the URL query string one by adding ?mode=xxxx where xxxx is the mode to use.

For winrm, only one mode can be used to require SSL access to remote: mode=ssl.

For example, the following command will process a winrm remote inventory over SSL (default port becomes 5986):

glpi-agent --remote=winrm://admin:pass@ --logger=stderr --tasks remoteinventory

For ssh, 3 modes are available:

  1. mode=perl can be set if perl is available on the remote to try using it for few specific cases (fqdn and domain),

  2. mode=ssh can be set to not try to use libssh2 for remote access,

  3. mode=libssh2 can be set to not try to use ssh command access if libssh2 fails.

You can combine modes. To do so, you just need to concatenate them using the underscore sign as separator: mode=perl_ssh or mode=ssh_perl are valid syntax

By default, the ssh mode is: mode=libssh2_ssh. So you don't need to specify both and they are still set if perl mode is set. libssh2 and ssh modes only need to be used if you have an issue with the other mode.

For example, the following command will process a ssh remote inventory using only libssh2 and enabling perl mode:

glpi-agent --remote=ssh://admin:pass@ --logger=stderr --tasks remoteinventory


As the inventory is run remotely, you may not obtain exactly the same inventory as if the agent was run locally.

For winrm remotes, the informations may miss or may differ from locally run inventory:

  • software installation date,

  • bios informations (as we can't run dmidecode),

  • devices name, type or description (mostly not localized via winrm),

  • databases services.

For ssh remotes, the informations may miss or may differ from locally run inventory:

  • printers,

  • databases services.