Chrome companion

The Chrome companion is a browser extension that connects intelligent.page to the browser you are already in. It does the two things no server can: capture a meeting as it happens, and read or act on a page behind your login. Install it once, and a task can bring back what a portal shows, fill a form on a site you have allowed, or turn a call into a transcript in the directory.

The companion talks only to intelligent.page. It has no other backend, it does not read pages you have not asked it to, and when it meets a login page, a two-factor prompt or a captcha it stops and hands the tab to you.

Capabilities

With the companion installed you can:

• Capture meetings: record Google Meet, Zoom on the web or Microsoft Teams on the web into a directory, with no bot joining the call
• Read the page you are on: pull the week's order status out of a portal with no export, or lift the numbers a dashboard will not let you download
• Work a queue behind a login: read and act on an internal tool as you, with your access, without handing over credentials
• Read many pages into one answer: twelve competitor pages into one comparison, brought back into the task
• Walk a flow and bring back the console: reproduce a checkout and return the errors and network activity
• Log the outcome where it belongs: what it did and what it found is written into the directory, not left in a tab

Input is delivered as genuine browser input rather than synthetic events, so pages behave exactly as they would with you at the keyboard.

Prerequisites

• Google Chrome, or a Chromium-based browser that accepts Chrome extensions
• An intelligent.page account signed in, in the same browser
• For meeting capture: a call running in a tab, and permission to use your microphone

Get started

Then ask for something that needs the browser. From the composer:

Open our supplier portal, read this week's order statuses, and put them in a sheet with anything overdue flagged.

The first time the task needs a site, the companion asks whether that site is allowed. Allow it, and the task proceeds in a visible tab.

Example workflows

Capture a meeting

Join the call, then click Record into intelligent.page in the extension. Audio is captured in short chunks and transcribed as the call proceeds, so a transcript with speaker names exists moments after the call ends — no bot in the participant list, nothing for the other side to admit.

Then ask:

From today's call with Northwind: what did we commit to, what did they commit to, and what contradicts what we believed last week?

See Capturing meetings.

Read a portal with no export

Go to the logistics dashboard, read the on-time delivery figures for each region for the last four weeks, and compare them with the targets in our operations doc.

The companion reads the page as you would see it and the task does the rest.

Compare pages across sites

Read the pricing pages of these five competitors and build a comparison of plans, prices and what each plan includes.

Each site is asked for separately; allow the ones you want read.

Reproduce a bug and bring back the evidence

Walk through our checkout as a new customer up to the payment step, and bring back the console errors and any failed requests.

Console and network activity come back with the page state, so the fix starts from evidence rather than a description.

Work a queue behind a login

In the support tool, go through the open tickets tagged "billing", summarise each, and draft a reply for the ones that are refund requests.

Reading proceeds under ask; drafting into the tool requires you to have allowed that site. Nothing is submitted without you.

Manage site permissions

Permission is per site, with three states, and the companion popup is where you set them.

Three rules hold regardless of what you set:

1. Acting requires an explicit allow. Reading may proceed under ask; clicking, typing and submitting never happen on a site you have not allowed.
2. There is no master switch. You cannot allow everything at once. Every site is decided on its own.
3. The blocklist beats your own approval. Banking, payments, webmail, health sites and government domains are refused even if you allowed them.

Change a site back to ask or block at any time; decisions persist until you change them. See Browser automation and safety for the full model.

Two more guarantees are worth knowing. The companion never picks the tab — it opens its own — and page text is evidence, never instructions: nothing a page says can change what the task was asked to do.

Troubleshooting

The extension does not see my session

1. Confirm you are signed in to intelligent.page in the same browser profile the extension is installed in.
2. Reload the intelligent.page tab, then reopen the extension popup.
3. Check the extension is enabled at chrome://extensions.

Recording produces no transcript

1. Check the browser has microphone permission for the meeting site.
2. Make sure the call is in a tab, not a desktop app — the companion captures tab audio.
3. Where the platform shows live captions, turn them on: speaker names are read from them.

A task stopped on a page

The companion stops for logins, two-factor prompts, captchas and account pickers, and hands you the tab. Complete the step yourself; the task continues from where it paused.

A site is refused even though I allowed it

It is on the blocklist — banking, payments, webmail, health or government. That is by design and cannot be overridden.

Managed install shows as "not installed"

The policy has not reached this browser yet. At chrome://policy, choose Reload policies, then check chrome://extensions.

See also

• Capturing meetings
• Browser automation and safety
• Platforms and integrations
• Data controls