AI Governance What it reads, and what it finds.

/ai-governance/sources Public beta

AI Governance reads the AI vendors, Microsoft 365, cloud platforms and security products you already pay for, then adds installs only where you need them.

[ 02 // connects to what you run ]
[ 02 ]

It connects to what you already run,
from a key to a computer.

You rip nothing out. StackJack reads the AI vendors, Microsoft 365, cloud platforms and security products you already pay for, and on Business it can propose changes in your AI vendor accounts for an administrator to approve. You do not need every source. Start with what you already pay for, add coding tools next, and add installs last. Every source finds AI use on every plan. Some need Pro or Business to scan text or to control AI use.

WayWhat it seesEffortPlan it needs
AI vendor admin accounts: Anthropic, OpenAI, ChatGPT Enterprise, GitHub CopilotUsage, cost and who did what in the vendor account. With an enterprise key it can also see chats and sessions.Low: paste a keyAny plan. The three admin connectors are included with a paid plan.
Microsoft 365 CopilotCopilot use by person and app, and Copilot conversations.Low: an administrator gives consentAny plan. Scanning conversations with rules needs Pro.
Azure OpenAI, Foundry and API Management, through an Event HubWhich model was called, when, by which application, and how many tokens.MediumAny plan
Gemini activity and conversations, Vertex AI, AWS BedrockGemini use in Google Workspace, Gemini conversations from Google Vault, and model calls made in your own Google Cloud or AWS account.MediumAny plan. Scanning text needs Pro.
Browser and gateway products: Island, Cloudflare AI Gateway and DefensXAI use seen by your browser or gateway product: facts only.Low: use a connection you may already haveAny plan
Security products: Microsoft Defender, DNSFilter, Cisco Umbrella, ThreatLocker, SentinelOne, CrowdStrike, Huntress, Bitdefender GravityZone and ESETAI apps and sites the product blocked or saw: inventory facts, never prompt text.Low: use a connection you already haveAny plan
StackJack’s own automation runs and tool callsYour automations’ runs and the calls through StackJack’s own tools. StackJack creates these sources.NoneAny plan. Scanning tool calls follows your plan.
Telemetry from coding tools: Claude Code, Codex, GitHub Copilot in an editor, Gemini CLIEach model call and tool call.MediumAny plan. Scanning text needs Pro.
Browser extensionThe AI websites staff use in Chrome, Edge and Firefox.High: roll it out to every browserAny plan finds sites. Pro and higher controls.
Endpoint agent (Windows)AI apps and tools on each computer, and the managed AI settings.High: roll it out to every computerAny plan finds AI. Pro and higher controls it.
Claude Inference hooksEvery Claude Enterprise request, before Claude acts on it.MediumBusiness
AI gatewayEvery call to OpenAI, Anthropic or Azure OpenAI that goes through StackJack.MediumBusiness
MCP proxyEvery tool call to a remote MCP server you approved.MediumBusiness
limits we say out loud
  • The endpoint agent runs on Windows.
  • The AI gateway does not serve Gemini CLI.
  • A Claude Team organization has no admin key to create. Use the coding-tool telemetry or the browser extension instead.
[ 03 // shadow ai discovery ]
[ 03 ]

Find the AI
nobody told you about.

The AI tool inventory lists every AI tool your organization uses. StackJack adds an item the first time it sees one. An administrator then approves or denies it. The inventory is the base of your AI register for clients, auditors and insurers.

The AI inventory: each AI tool with its kind, whether it is approved, how many people used it and when it was last seen.
Every AI tool found, how many people use it, and whether anyone approved it.
1
one list

Everything AI, in one inventory

Coding tools, AI web apps, models and model deployments, API keys, IDE and browser extensions, MCP servers, agents and built-in AI features.

Items come from the activity StackJack receives, your vendors’ admin consoles, Microsoft 365, the browser and network products you connected, a request from your staff, or you.

every plan
2
a decision for each

Approve it or deny it

A new item starts as New, needs review and raises an “Unapproved AI tool” alert. An administrator approves or denies it. Use of a denied item raises a “Denied AI tool in use” alert.

StackJack builds no score or ranking of people from this.

3
staff can ask

A queue for tool requests

Anyone on your staff can ask for an AI tool or an MCP server, with a reason. An administrator decides with one click. The person who asked is told the decision.

every planApprove from Teams: Pro
4
the record

What each item records

Vendor, owner, where it was seen, the contract you hold with the vendor, whether training on your data is switched off, the kinds of data it handles, who used it in the last 30 days and an estimated cost when a figure exists.

AI can suggest an EU AI Act risk class. It is labelled as a suggestion until a person confirms it.

5
act on it

From a finding to a fix

The Act menu on an inventory item, an alert or a finding opens a ticket in your PSA, asks the approvers for a decision, or proposes a block of a web app’s domain in DNSFilter, Cloudflare or Cisco Umbrella.

ticket and decision: Proapply fix: Business

A fix only proposes. An administrator types a confirmation phrase, with a second factor, before anything changes.

[ 04 // browser extension & endpoint agent ]
[ 04 ]

Need to see more?
Add the extension and the agent.

Most of the picture comes from the connections above. For those who need the rest, two installs see what a vendor report cannot. Both find AI on every plan. On Pro and higher both also control it. You roll them out with the tools you already use to manage computers. Your name and icon can replace ours on the agent.

The coverage page: one row per computer or person, showing the endpoint agent, the security product and the browser extension, and whether each is reporting.
Coverage: which computers and people each install reaches, and whether it reports. Names replaced.
browser extension

Chrome, Edge and Firefox

finds sites: every plancontrols: Pro and higher
On every plan
  • Finds the AI websites your staff use. Each visit appears in your inventory, so you can approve or deny it.
  • Shows coverage: which computers have the extension, by browser and version, and when each last reported.
  • Alerts you when it can no longer read an AI site.
On Pro and higher
  • Allows, warns about or blocks AI sites you have not approved.
  • Allows, warns about or blocks personal accounts on AI sites.
  • Checks each prompt, paste and plain-text file against your data-loss rules, and can allow, warn or block it.
  • Masks text in the prompt box on a site StackJack has marked safe to change. Where masking is not possible, your mask fallback setting applies, which is a block unless you change it.
  • Can record the AI’s answers, at a capture level that keeps text.
What it never does
  • It reports a visit only to a site on StackJack’s reviewed list of AI sites, or on your approved or denied list. It never reports other browsing.
  • It never stores a prompt, a paste or a file on the computer.
  • If it cannot reach StackJack, it lets the send through and keeps a note, with no text, that the send was not checked.

It reads the page, to check sends and accounts, on ChatGPT, Claude, Gemini, Microsoft Copilot, Perplexity, DeepSeek and Grok. Install it with Intune, Group Policy or one PowerShell script. Edge uses the Chrome Web Store listing.

endpoint agent

A small service on each Windows computer

finds AI: every plancontrols: Pro and higher
On every plan
  • Finds installed AI apps, AI browser extensions, the local MCP servers that tools such as Claude Code, Cursor, Codex, Claude Desktop and Gemini CLI start, and the AI sites the computer looked up.
  • Keeps your managed AI settings in place, and puts them back when someone changes them.
  • Reports the computer name and Windows version, so you can tell computers apart.
On Pro and higher
  • Local AI tool control: your approvals and data rules apply to each tool call that Claude Code, Cursor, Codex, Claude Desktop and Gemini CLI make.
  • The paste check, with the same data-loss rules as everywhere else.
  • AI app control and AI site control: report, warn or block the ones you have not approved.
  • Your own name, icon and tray text, and your own package signature.
What it never does
  • It does not read web pages, open encrypted traffic or act as a proxy. Only the browser extension reads text inside a page.
  • It uses no driver and does not log keystrokes. The paste check watches the two paste shortcuts only, and only while an AI app is in front.
  • It does not scan files, and it does not stop AI work when StackJack is slow: it lets the action through and records that it was not checked.

Install it with your RMM, Intune or Group Policy as a signed Windows installer. An uninstall code can lock removal. Each agent checks the signature of an update before it runs.

[ 05 // faq ]
[ 05 ]

Questions.

Do I have to install anything?

Not for the scan, and not for most sources, which use connections and keys you already have. The browser extension and the Windows endpoint agent are there for those who need to see more.

Does the browser extension read all of my browsing?

No. It reports a visit only to a site on StackJack’s reviewed list of AI sites, or on your approved or denied list. It never reports any other browsing.

Does the endpoint agent log keystrokes?

No. It is an ordinary Windows service with no driver. The paste check watches the two paste shortcuts only, and only while an AI app is the window in front.

Does AI Governance collect anything when I turn it on?

No. A new organization collects nothing until you add a source. For staff in the EU or the UK, collection also waits until you record how you told your staff.

[ 07 // start ]

Know what AI is doing.

AI Governance is in public beta. Start free, add one source, and see which AI tools your people use.