How Do You Deploy an AI Governance Agent Through an RMM?
Give the RMM one silent install command, the organization key and a client ID for each client. Run it on a small pilot group first, add a health check as a script monitor, then roll it out to the rest of the client's devices. Each step below says how long it takes and how to check that it worked.
The same steps work in any RMM that can run a script on a device and read its exit code. There is no plugin and no file to attach on Windows, macOS or Linux.
The Six Steps
- Get the organization key and the client ID (2 minutes). The key is one per service provider and is a secret. The client ID picks which client a device joins and is not secret. Both are on the Deploy page in the console. Check: the console checks the pair before anything is installed, and setup shows the client and the provider by name, so a typo is caught before a device joins the wrong client.
- Store them in the RMM (5 minutes). Keep the organization key as a secure or global variable. Keep each client's ID in a per-site custom field, so one script serves every client. Check: two different clients' sites show two different client IDs in the RMM.
- Create the install script (10 minutes). Use the commands in the next section, with the variables from step 2. A silent install accepts the agent installation terms on your behalf as the provider, which confirms that the client has authorized the install, as the Partner Agreement requires. Check: the script saves in the RMM library and shows your variables in place of the placeholders.
- Pilot on two to five devices (15 minutes). Use your own shop or a friendly client. Check: each pilot device appears under the client's Devices page in the console within minutes. New clients start in Passive mode, which records AI use and blocks nothing.
- Add the health check (10 minutes). Run
proctoredai-agent checkas a script monitor, for example every 15 minutes. It exits 0 when the device is protected, 1 when the agent runs but a check fails, and 2 when the agent is not installed or not running. Add--jsonto fill custom fields. Check: stop the agent on a test device and confirm the RMM raises an alert. - Roll out to the rest of the client. Run the same script on the client's remaining devices. Running it again on an enrolled device re-enrolls it and retires the old record, so a repeat run never leaves a duplicate. On Windows, if the client has an uninstall code, reinstalling over an existing install needs it. Check: the client's setup checklist in the console shows All Devices Ready.
The Install Commands
Run these as administrator or root. The Deploy page fills in your key and the selected client's ID; the examples use placeholders.
Windows (PowerShell script type)
One line downloads the agent from the console:
[Net.ServicePointManager]::SecurityProtocol = [Net.ServicePointManager]::SecurityProtocol -bor [Net.SecurityProtocolType]::Tls12
& ([scriptblock]::Create((Invoke-WebRequest -UseBasicParsing 'https://proctoredai.com/install.ps1').Content)) -OrgKey 'GOV-XXXXX-XXXXX-XXXXX-XXXXX' -ClientId 'CLIENT-ID' -AcceptTerms
macOS and Linux
curl -fsSL "https://proctoredai.com/install.sh" | sudo sh -s -- --org-key GOV-XXXXX-XXXXX-XXXXX-XXXXX --client-id CLIENT-ID --accept-terms
One installer serves every client. Updates install silently afterward, and each one is checked against ProctoredAI's signed release key before it installs.
Lock It Down After the Pilot
- Uninstall protection. Set an uninstall code per client in Client Settings. Every uninstall path then asks for it, and a wrong code raises an alert.
- MDM settings. Where the client has an MDM, push the proxy and trust settings so people without administrator rights cannot change them. On Windows and macOS the agent also puts its own settings back within 30 seconds if anyone removes them.
- MDM-managed installs. Where a client's Intune or Jamf Pro is connected, the console lists the computers that are missing the agent and can install or repair it through the MDM.
Chromebooks
ChromeOS does not allow native agents. Chromebooks run ProctoredAI as a managed browser extension, force-installed from Google Admin with the managed configuration shown on the Deploy page. It applies the same policy, but does not cover everything the desktop agent does.
What to Plan For
- Devices that are off. A device enrolls the next time it is on and online, so the RMM's own retry settings matter.
- Authorization first. The agent inspects AI traffic on the device. Agree the scope with the client and tell staff before you deploy, whichever mode you choose.
- Rotating the key. Enrolled devices keep working after a key is rotated, but update the RMM script, which holds the old key.
Questions
Which RMM tools does this work with?
Any RMM that can run a script on a device and read its exit code. The install and the health check are plain commands with no RMM-specific plugin. See the integrations page for PSA, chat, MDM, identity and SIEM connections.
Does the agent need a browser extension?
Not on Windows, macOS or Linux. The agent runs on the device and routes AI traffic through itself, so it covers every browser, desktop AI apps and command-line tools. Chromebooks use a managed extension because they cannot run a native agent.
How long does a client take?
Setting up the script takes under an hour once. After that the agent installs from the RMM in minutes per device, and most clients are observed for two weeks before any rule blocks anything.
Can I deploy without a pilot?
Yes, but a pilot of two to five devices catches a wrong client ID or a policy surprise before it reaches a whole client. Passive mode means nothing is blocked while you check.
How do I know an agent stopped?
The health check exits 1 or 2, and the console shows the device as Not Enforcing or Not Reporting and raises an alert, which can open a ticket in your PSA.