Skip to content

Asking your assistant about the page you're on

How to get the Chrome side panel to read the page in front of you, exactly what it takes from that page, and what it never sees.

Last updated August 27, 2026

On this page

The Praxivara side panel can read the page you are looking at, but only when you ask it something that needs the page. There is nothing running in the background, no script sitting on every site you visit, and no record being built of where you go. The rest of the time it is just your assistant, in a panel.

How to ask

You do not need a special command. Ask the way you would ask a person sitting next to you:

  • "Summarise this page."
  • "What does this clause actually mean?"
  • "Pull the numbers out of this table."
  • "Is there anything in here I should push back on?"

There are shortcuts too:

  • The Summarise this page and Key facts chips on an empty conversation.
  • Right-click anywhere on the page and choose Ask Praxivara about this page. The panel opens with an opening line already typed, ready for you to edit or send.
  • Select some text first, then right-click and choose Ask Praxivara about this. The selection is dropped into the message box so you can say what you want done with it.
  • Type /page in the composer for a prompt that starts "Read this page and…".

When the assistant decides it needs the page, it fetches it and carries on with your question in the same breath. You may briefly see it working; you do not have to do anything.

What it takes from the page

It takes a snapshot of the page as it stands at the moment you ask — the whole page, not just the part you can see:

  • The visible text of the page, up to roughly 60,000 characters. Longer pages are cut short, and the assistant is told they were.
  • The page title and its web address.
  • Anything you had highlighted, if you had highlighted something.

What it does not see

  • It cannot act. It reads and it talks. It cannot click a button, type into a box, submit a form, or change anything at all on a page.
  • Things Chrome is not drawing are skipped. Scripts, styles, and elements Chrome is not rendering — a collapsed panel, a closed tab strip — are left out. Text that is merely hard to see, such as text positioned off the edge of the page, is still read.
  • Plain form fields are skipped. Text you have typed into an ordinary input box or text area is not collected. Text in a rich editor — an email you are composing, a document you are writing in — reads as ordinary page text and is collected, so treat an unsent draft as part of the page.
  • Images are not read. It reads text. If you want it to look at a picture, take a screenshot and attach it to the conversation instead.
  • Other tabs are not read. Only the tab you are on when you ask, and only that one time. Ask again a minute later and it reads again, fresh.
  • Nothing is read in the background. A page you never ask about is never opened by the extension.

Pages it cannot read at all

Chrome itself blocks extensions from a few places, and the assistant will tell you so rather than quietly returning nothing:

  • Chrome's own pages — settings, history, the extensions page, anything starting chrome://.
  • Extension pages and view-source pages.
  • Files opened straight off your computer, starting file://.
  • The Chrome Web Store.

You may also see a message if the tab has not finished loading, or if the page genuinely has no readable text — a PDF in Chrome's viewer is the usual example. If it is a PDF you have on file, attach it to the conversation instead and the assistant will read it properly.

If it asks for permission first

Sometimes a small card appears asking to read the site you are on, with Allow and Not now. That is Chrome's design, not a Praxivara one: the extension is not given access to other websites when you install it, so the first time the assistant needs to read a particular site, Chrome asks. Your answer is remembered for that site, and you can change or remove it later under Extensions, Praxivara Assistant, Site access. Saying no is fine — the assistant simply carries on without the page.

A page cannot give your assistant orders

Web pages sometimes contain hidden text written to steer an AI assistant into doing something on the reader's behalf. Everything the panel reads is handed over clearly labelled as information to read, never as instructions to follow. If a page tries it, the assistant is told to mention it to you rather than act on it.

It is still worth a moment's thought before asking about a page. When it reads a page, the text goes to your own Praxivara account so the assistant can answer that one question — so treat it the way you would treat pasting the page into a chat. On a bank statement, a medical record or anything you would not paste, it is reasonable to say no.

Was this page helpful?
Still need help? Contact support