Skip to main content
Connect your AI

The StackJack endpoint agent

The StackJack endpoint agent is a small program that runs on each Windows computer you manage. It finds the AI apps and AI tools on the computer and keeps your managed AI settings in place. On the Pro…

Written By Christopher Scaminaci

Last updated 13 days ago

The StackJack endpoint agent is a small program that runs on each Windows computer you manage. It finds the AI apps and AI tools on the computer and keeps your managed AI settings in place. On the Pro plan and higher it also controls local AI tools, checks text that people paste into AI apps, and stops the AI apps and AI sites you have not approved. Your name and icon can replace ours on everything the agent shows.

At a glance

  • What it does: finds AI apps and tools on Windows computers and keeps your managed AI settings in place. On Pro and higher it also controls them.
  • Plan: every plan finds AI. Control needs the Pro plan or higher.
  • Who: an Administrator.
  • Where you set it up: AI Governance → Staff computers → Endpoint agent.
  • Where you see the data: Activity → Activity explorer, AI inventory → Inventory and the Devices tab of the same page. Staff computers → Coverage lists the same computers beside the browser extension and your security products, and says which of them are sending data.

This release runs on Windows only. This page says what the agent does, what it never does, and what your staff see. The setup is in Installing the endpoint agent. The choices you make are in Endpoint agent policy.

What it does on every plan

  • It finds AI on the computer. It lists the AI apps that are installed, the AI browser extensions, the local AI tool servers (MCP servers) that tools such as Claude Code, Cursor, Codex, Claude Desktop and Gemini CLI start, the security products it recognizes, and the AI sites the computer looked up recently. It sends these as inventory facts. A fact about an installed app does not count a person toward your staff count.
  • It keeps your managed settings in place. The settings StackJack makes for your AI tools are checked on each computer and put back when someone changes them.
  • It says which computer it is. With each report the agent sends the computer's name (the Windows computer name), its Windows version and build (for example "Windows 11 Pro 24H2 (26100.2033)"), and when it last read its policy. They label the row on the Devices tab, so an administrator or an auditor can tell the computers apart. The computer name is shown as it is. The person is shown the way the rest of AI Governance shows people, so under the pseudonymous setting it stays a pseudonym until a name reveal is approved. A computer running an older agent shows "Needs a newer agent" until it updates.

What it does on the Pro plan and higher

  • Local AI tool control. The agent can watch the tool calls that Claude Code, Cursor, Codex, Claude Desktop and Gemini CLI make, and apply your approvals and your data rules to them. See Local AI tool control.
  • The paste check. When a person pastes into an AI app, the agent checks the text with the same rules as everywhere else in AI Governance.
  • AI app control. An AI app you have not approved is reported, flagged to the person, or ended within about a second of opening, as you choose.
  • AI site control. An AI site you have not approved is reported, flagged to the person, or stopped on the computer, as you choose.
  • Your own brand. The name, icon, tray text and the package signature can be yours. See the Branding tab of the Endpoint agent page.

On a lower plan the agent finds AI and keeps settings, and it enforces nothing.

What it never does

  • It does not read web pages. It does not open encrypted traffic and it is not a proxy. Text inside a web page is read only by the browser extension, which the agent can install and help.
  • It uses no driver. It is an ordinary Windows service and a small helper for the person who is signed in.
  • It does not log keystrokes. The paste check watches for the two paste shortcuts only (Ctrl+V and Shift+Insert), and only while an AI app is the window in front. It reads no other key. A paste through a right-click menu or by drag and drop is not checked.
  • It does not scan files. File contents are outside this release.
  • It does not send text when you chose not to. At the facts-only capture level no text leaves the computer. See Capture levels and what your employer sees.
  • It does not stop AI work when StackJack is slow or unreachable. It lets the action through and records that it was not checked. You can choose to block instead on the policy page.
  • It does not prove who a person is. The sign-in name it reports is the Windows sign-in, and nothing proves it. StackJack marks it "not verified".

What your staff see

Each computer shows a branded tray item with your product name. Its menu shows what the agent watches, in your words if you set the tray text, and a link to your support page if you set one. There is no exit item, so a person cannot close the agent from the tray. When the agent ends an AI app or stops a site, the person sees a short notice with your block message.

The staff page explains all of this in words you can hand to your staff.

How it works with the browser extension

The agent and the browser extension complete each other. The agent installs and configures the extension. It also tells the extension which Windows person is signed in, and the extension shows that name as "not verified". Only the extension reads text inside a web page.

Updates

By default each agent updates itself from the StackJack download host, and checks the signature of each new version before it runs it. You can hold every agent of an organization at one version, or switch an organization to RMM only, so that agents never update themselves and you push each new package yourself. See Endpoint agent policy.

Removing the agent

When you set an uninstall code, stopping or removing the agent needs the code. Replacing the package with a newer version is an upgrade, not a removal, and needs no code. A device where the agent was stopped without the code shows as tamper suspected. A person with administrator rights on the computer can still end the process by force, so the agent reports its own absence to you. See Installing the endpoint agent.

Where to go next