The Waxell Browser Extension
Most AI use happens in a browser tab. Waxell captures and governs it with the Waxell AI Monitor extension, which the Waxell agent force-installs into each managed browser. The extension reads the conversation in the page itself, so it sees the prompt and the full reply without decrypting any network traffic.
Supported browsers
| Browser | How it is installed | Extension id |
|---|---|---|
| Google Chrome | Force-installed from the Chrome Web Store | npipfngnlpinjdbbdoihdoihfhpmibef |
| Microsoft Edge | Force-installed from Microsoft Edge Add-ons | egoiaobngfnpdbnpkeofaogmfohpoffc |
| Mozilla Firefox | Force-installed from a signed package Waxell hosts | waxell-ai-monitor@waxell.dev |
Chrome and Edge use different ids because managed Edge will not force-install a Chrome Web Store item. If you allow-list extensions in your browser policy, allow the id for each browser you use.
On Windows, the Waxell installer writes the browser policies for you
(ExtensionInstallForcelist, ExtensionSettings, and the Firefox enterprise
policy) when it installs, so an MSI deployed through Intune or an RMM is all you
need.
On macOS, the browser policies come from the browser profile you download from the Setup page and push through your MDM alongside the Waxell profile. See macOS deployment.
Sites covered out of the box
- ChatGPT —
chatgpt.com,chat.openai.com - Claude —
claude.ai - Gemini —
gemini.google.com - Microsoft Copilot —
copilot.microsoft.com,copilot.com,copilot.cloud.microsoft,m365.cloud.microsoft - Perplexity —
www.perplexity.ai
To govern another AI site, add its hostname to Browser extra hosts on the Guard's Browser Capture tab. The extension starts covering it on the next policy update (devices check for policy about once a minute), with no extension release. On macOS, re-download and re-push the browser profile after changing this list so the browser grants the new site.
How the extension recognises a send, the reply and the account on each site is defined by site adapters. Waxell maintains them; you can override one for your workspace under Endpoints → Site Adapters.
How a capture reaches your workspace
- The extension records each exchange in the page: your prompt, the model's reply, the model, and which account you were signed in with.
- It hands the record to the Waxell agent on the same machine over the browser's
native-messaging channel (
ai.waxell.bridge). The agent writes it to a local spool before acknowledging, so nothing is lost if the browser closes. - The agent redacts secrets on the device (see Privacy & DLP), stamps the record with the device and user, and uploads it.
- The capture appears in Conversations and Device Activity within seconds.
The same channel carries policy the other way: when you change the Guard, the agent pushes the new policy to the extension, which applies it on the page.
Turning upload off
Browser capture on the Browser Capture tab decides whether captures leave the device. It is on by default. Turn it off on any Guard layer (the whole workspace, a group, a single device) and the agent holds captures on the device instead of uploading them. Turning it back on delivers what was held. The device shows held on the Health page while upload is off.
Personal and corporate accounts
Every capture is labelled with the account it was sent from: corporate, personal, or unknown. ChatGPT and Claude are classified directly from the page. Gemini, Perplexity and Copilot are classified by the signed-in email, so add your company's email domains to Corporate email domains or those sites read as personal.
The account label is on every capture today, and Captures and Fleet can be filtered by it. Personal AI mode, which will warn or block sends from personal accounts, is shown as Coming soon in the Guard editor and cannot be turned on yet.
Confirming the extension is working
The Health page shows a Capture state for every device:
| State | Meaning |
|---|---|
| on | The extension is installed and talking to the agent. |
| held | Working, but Browser capture is off, so captures stay on the device. |
| on · Edge uncovered | Working in some browsers, but one with encrypted DNS has not reported in. |
| no ext | The agent has never heard from the extension in a browser that needs it. |
| dark | The extension used to report but has been silent for a day. |
| n/a | No browser AI activity seen yet. |
Each device reports the extension once per browser (Chrome, Edge, Firefox) the first time a capture flows through it.
If a device shows "no ext"
The agent has not received anything from the extension. Check, in order:
- Is the extension installed and enabled? Open
chrome://extensionsoredge://extensions. It should list Waxell AI Monitor as installed by your organisation, with the id from the table above. - Is the native-messaging host registered? On Windows, the key
HKLM\SOFTWARE\Google\Chrome\NativeMessagingHosts\ai.waxell.bridge(and theMicrosoft\Edgeequivalent) should exist. Reinstalling the agent rewrites it. - Is an application-control tool blocking the bridge? Browsers start
WaxellBrowserBridge.exe(Windows; it is launched throughcmd.exe) or the Waxell BrowserBridge (macOS) every time the extension connects. Tools that allow-list executables — ThreatLocker, AppLocker, WDAC and similar — must allow it to run as a child of the browser, not only allow the installer. On Windows it lives atC:\Program Files\Waxell\Bridge\WaxellBrowserBridge.exeand is signed by Waxell. - Has anyone used AI in that browser since install? The state updates on the first capture.
Without the Waxell agent
If the extension is installed but the Waxell agent is not, the extension uploads captures directly (redacting secrets itself first). Policy only reaches the extension through the agent, so in this mode it records but does not enforce Browser DLP or Personal AI mode. Install the agent to enforce.