Frequently asked questions

Installing

Why does Chrome say the extension is not trusted by Enhanced Safe Browsing?

With Enhanced Safe Browsing on, Chrome shows that line for extensions from developers who are new to the Chrome Web Store, until they have followed its policies for a few months. It is not a finding about the extension, and it goes away on its own.

Where is the icon after I install it?

In the Extensions menu, the puzzle piece beside the address bar: Chrome and Edge do not put a new extension on the toolbar. Open the menu and pin Page Scanner (in Edge, choose "Show in toolbar"). The welcome page the install opens says the same, and says when the icon is on the toolbar. Without pinning it, Alt+Shift+P opens it on any page, unless another extension already has those keys; then set others at chrome://extensions/shortcuts.

Why did a page open when I removed the extension?

A copy from the Chrome Web Store or Edge Add-ons asks the browser to open one page on pagescanner.app after it is removed. The page asks why, with ten boxes to tick and a Send button, and has no text box; pressing Send records which boxes you ticked in Google Analytics, and closing the tab records nothing. With Report usage statistics off, its address carries no version, browser or store. The privacy policy says what the page sends.

Scanning

Which pages can it scan?

http, https and file pages. Chrome reserves chrome:// pages, the Chrome Web Store, other extensions' pages (the built-in PDF viewer is one), about: and view-source: for itself, and a new tab has nothing to scan. On those the panel offers Open a PDF or image instead.

What is the bar that appears at the top of the browser during a scan?

Chrome's notice that a debugger is attached to the tab, which is how the extension lays the page out for a vector PDF. It shows for the length of the scan and about five seconds after, and Chrome does not let an extension hide it. Why the debugger says what it means.

Why is my PDF an image instead of text?

DevTools was open on the tab. Only one debugger can attach to a tab at a time, so the scan used the raster fallback, which scrolls and stitches screenshots. Close DevTools and scan again. The fallback scrolls the window only, so on an app that scrolls inside a box, as Gmail does, it gets one screenful; the normal vector capture gets all of it.

How long a page can it capture?

60,000 CSS pixels on each side by default. Maximum page height in Settings raises the height to 120,000 or 240,000. A page longer than that is captured up to the limit, and the export message says how much was left out. Past 60,000 pixels the PDF keeps its text and sharpness, the editor's preview and a PNG or JPEG made from it get fewer pixels per part of the page, and a single long page export comes out as several long pages.

A page that draws only what is on screen, such as a long ChatGPT conversation, is captured up to 10 million pixels whatever the setting, because below the limit it would otherwise be blank. The PDF of such a page is as sharp as any other; the editor's preview gets coarser the longer it is. A page that is simply long, a long Gemini or Claude conversation among them, stops at the setting.

Does it open collapsed sections before capturing?

Turn on Open folded sections before capture in Settings and the scan opens what marks itself folded, then folds it back. A section behind a "Show more" button stays as it is, and an image that takes more than four seconds to load is captured as far as it got.

What does the clean-up remove?

What you tick in Settings: ads, cookie and consent banners, chat widgets, and pop-ups and overlays, as far as its rules recognize them. What it hides is not in the file. The consent-banner words it reads are English, Korean, Japanese, German and Italian. For a capture kept as a record of how the page looked, untick all four.

Can I capture several tabs into one file?

Two to ten tabs from one window. If one of them cannot be captured, the scan stops rather than leave a gap in the file. A PDF stamps each tab with its own address; an image carries the first tab's.

Can it capture a page together with the terms it links to?

Yes, as several tabs: open the linked pages in tabs, select them with the page, and scan them into one file with the stamp and an integrity record. A page and its terms in one file has the steps. A scan does not follow the links itself; it captures only the tabs you select.

Can it capture every page of a site?

Up to 30 of them, with Map this site in the popup: it lists the site's pages, one of each kind, captures them in background tabs once you press Start, and draws them as a flow joined by their links. It stays within the site, never opens a link that would sign you out or delete something, and keeps to the site's robots.txt. Map a site has the steps.

Why does a site map ask to read the site?

To list a site's pages before capturing any, Page Scanner reads them, and Chrome lets an extension do that only for a site you allow. It asks for the one site you are mapping, when you press Find pages, and for no other. The pages go nowhere but your browser. Chrome keeps the access until you remove it on Page Scanner's details page under Site access.

Can it make a step-by-step guide of a task?

Yes, with Record steps in the popup: click through the task once, stop, and each click that changed the screen is a step with its picture, a number on what you clicked and a sentence you can rewrite. Sign-in and ID fields are covered in the pictures. The guide saves as one PDF, a page per step. Record steps has the details.

Does a recording keep my password?

No. What you type is not stored on its own, and password, sign-in, one-time code, card and ID number fields are covered with a black box in every picture, which stays black in the PDF. A field of another kind shows what you typed in it, as your screen did, so look through the pictures before you share a guide. Everything stays in your browser.

Can it record how a page scrolls, its animations included?

Yes, with Scan scrolling in the popup, as a storyboard rather than a video: it scrolls the page by itself and cuts it into scenes, each with six frames from its start to its end, what moved said in words and how it feels when you stop scrolling. You can play it back in a flipbook, mark the frames up, add a note for each scene, and save it as one PDF to hand to an AI coding agent. Scan scrolling has the steps, and an agent can ask for the same storyboard with page-scanner storyboard. It makes no video or GIF: a video is large, and it shows what moves without saying what it is, by how much or over how much scroll, which is what rebuilding it takes.

Why did my storyboard come out as one long scene, or stop early?

It scrolls the page itself, so an app whose content scrolls inside a box of its own (a mail client, a chat) shows as one screen that holds. A scan takes two minutes at most, which is about a hundred screens; past that it stops and says so. Scrolling, clicking or typing on the page while it scans stops it too, and what was scanned so far opens.

Does the phone width show the mobile site?

It lays the page out 390 px wide, so the site's responsive CSS responds. A site that picks its mobile version from the browser's name or from touch support shows its desktop version.

Does forcing light or dark mode work on every site?

On sites that read the color scheme from CSS, and on sites that mark it with a dark class or a data-theme on the page, even after you picked the other theme on the site. A site that picks its theme in JavaScript some other way when it loads keeps the one it chose.

The scan pins anything sized to the screen at its on-screen height before laying the page out. A height the page sets inline with !important, or in a stylesheet served from another domain, is out of its reach.

Why does the bottom of a long page sit a few pixels off?

On a Retina screen Chrome draws the page at full resolution and prints at a lower one, and rounds some sizes differently for each. The scan corrects the largest of these, line spacing, so what is left is a fraction of a pixel per paragraph. A line with an emoji or another alphabet in it may sit a little closer to its neighbors in the capture.

How are videos, canvases and photos captured?

A video is a still: the frame it paused on, or an empty area, as Settings choose. Text a map or chart paints into a <canvas> is a picture, which a PDF export can recognize on your computer and lay in as searchable text; see Export. A photo is as sharp as your screen shows it.

Does the page's own print stylesheet apply?

No. The scan switches off a page's print page breaks and margins so the PDF matches the screen, and puts them back afterwards. A margin or paper size the page forces with !important still gets through.

Exporting

Are links kept?

In a PDF, yes, including links that jump within the page. A PNG or JPEG is a picture and has none. A link under a blur, a redaction or an erase is left out, since its address could give away what the mark hides, and so is a link that runs a script.

Why did a redaction turn my PDF into an image?

A redaction removes the text under it and keeps the rest of the page as text. When it covers a picture, or a drawing that runs past its edge, the only way to cut that out is to make the page an image, and the editor says so before you extract. See the editor guide.

How do I get the whole page as one long PDF page?

Choose Single long page under Page layout in the Output Format panel: one page the exact size of the crop, with no margin, no scaling and no page breaks. A crop taller than 60,000 pixels comes out as several long pages, none taller than that, and some PDF viewers cut off a page past 200 inches, which the panel warns about. See Page layout.

Can the page breaks avoid cutting lines in half?

They do by default. On A4, Letter or phone screen pages, Where the text allows moves each break up, by at most an eighth of a sheet, to the widest gap between lines it finds there, which can cost one extra sheet. A page with no gap to cut at, such as a full-width picture, keeps the plain break, and Every sheet the same turns the moving off. See Where the page breaks fall.

Why is the text small on A4 or Letter?

A wide page is scaled to fit the sheet, so a 1920 px layout has small body text on paper. The Output Format panel shows the resulting size in points before you export, and a single long page is not scaled. Narrowing the page to the sheet goes as far as the page's CSS allows.

How long are captures kept?

Up to 12 hours: a capture older than that is deleted the next time the extension starts. Export what you want to keep.

Which files can I open?

A PDF, a PNG, a JPEG or a WebP. A picture can be up to 32,767 pixels on a side and 90 million pixels. A password-protected PDF cannot be opened, a PDF's own comments, highlights and links are not carried over, and a PDF past about 51 A4 pages opens on its first pages.

AI features

Why does AI Scan say there is no model?

AI Scan and a tidied export ask Chrome's built-in model, which Chrome offers only on computers that meet its requirements, and Edge only in its preview channels. Where there is none, the panel offers a small model to download and run in the browser, with a button that names it and its size; press it, and AI Scan and tidying use that model once it is downloaded. You can also choose a model you run yourself, with Ollama, LM Studio or llama.cpp, in Settings › AI Model. Pattern Scan works either way.

Can I use Ollama, LM Studio or another local server?

Yes: pick it, marked (Local), under Model Server in Settings › AI Model, check the port, and pick one of its models. Chrome asks once to let Page Scanner reach localhost, and the text goes to that server on your computer and nowhere else. The server has to be running, with a model in it; Page Scanner starts neither.

Can AI Scan run without Chrome's model or a server?

Yes: WebGPU in Settings › AI Model downloads a small model from Hugging Face, once, and runs it on your computer's graphics card through WebGPU. Page Scanner recommends one for the graphics card it finds. It is slower than Chrome's model and finds less, so check what it found before you hide it.

Does the text leave my computer?

Not for AI Scan or tidying. Chrome's model runs inside Chrome, a local server runs on your computer at localhost, and a model downloaded into the browser runs in the browser. The download is the model's own files; nothing of yours goes with it. What can leave is what you say to the Assistant or the side panel, with the choices and counts they work from, as the next question says.

Does the Assistant send what I type anywhere?

Every message you send it goes to Page Scanner's server, where a model reads it; nothing is read on your computer. With it go the conversation so far, the editor's choices, and counts such as how many email addresses Sensitive Text listed. Never the page's text, pictures, address or title. The side panel does the same with the scan's choices and what it counts on the page (how many ads, whether there is a dark version). Page Scanner's server in Settings › AI Model switches both off, and the privacy policy says what the server keeps.

Can the Assistant see the page?

It sees the editor's choices and counts: the format, the crop, whether a main content was found, how many sensitive finds and marks there are. It never gets the page's text or pictures, so it can tell you what the file will be ("what format will it be?") but not what the page says, and it cannot summarize it. To take the page's words somewhere, export it as Markdown or Word.

Why is the page narrower when I scan from the side panel?

The side panel takes its width from the window, and The window, as it is captures the page at the width it has on screen, which is the width left beside the panel. Say "A4" or "Letter" in the panel for a page laid out for paper, or close the panel and scan from the popup for the full window.

Can I capture a page I translated in Chrome?

Yes. Translate it first with Chrome's own translate, then scan: the capture keeps the translated text as real, searchable text. Every file made from it says it is a machine translation (the PDF's Subject, a Markdown file's front matter, a Word file's description, the reference file), so it is not quoted later as the page's own words. The export page lists where.

Where did the Assistant go?

The Assistant is a preview, and Page Scanner can close it, or only its server, without an update. The editor asks when it opens and every ten minutes while it stays open; a closed Assistant has no icon on the rail, and with a closed server the Assistant reads nothing and says so. It comes back the same way. The side panel is a preview too: closed, the popup has no button for it, and an open panel says so. So is the site map: closed, the popup has no Map this site, and a map already made still opens. And so is Record steps: closed, the popup has no Record steps, and a recording already made still opens. And so is Scan scrolling: closed, the popup has no Scan scrolling, a scan already running can still be stopped, and a storyboard already made still opens.

Why does the Assistant say this month's requests are used?

Page Scanner's server reads 10 messages a month for each install, free, shared by the Assistant and the side panel. A message counts once, however many steps it takes, and one that failed before the model read it was not counted. The first line of each answer says how many are left, and once they are used a red note at the end of the conversation says when they come back. Until then the Assistant cannot read a message; the panels and the toolbar do everything it does, and the popup scans with the settings.

Why did the Assistant not understand my request?

The Assistant has a model on Page Scanner's server read every message. When a request is unclear it asks what you meant, so answer it in the next message. When it cannot do what you asked, it says so and where in the editor it is done by hand: it does not draw marks, move the crop by a number, or pick colors, the stamp's time, a brand, a watermark or a preset, and it cannot answer questions about what the page says. Naming what you want helps: a format, a paper size, the stamp, the crop, a tool or a way to hide text. A message that used all 8 of its steps stops and says so; say what is left in the next one.

Agents and the command line

Why does the debugger bar appear when an agent scans?

For the same reason as a scan you start. When an agent runs unattended it will appear without you having clicked anything.

Can an agent scan a tab with DevTools open?

Close DevTools first, or scan another tab. Agent scans use the vector capture only, and report an error rather than hand back a picture of the page.

Can an agent crop the capture?

It always captures the whole page, since it has no preview to crop against.

What happens with a very tall page?

The same height limit applies, and the result sets truncated so the agent knows the file is missing the rest rather than passing it off as the whole page.

Can it watch a page for changes?

With the command line and your system's scheduler: scan captures the page's text, diff compares it with the last capture and exits 1 when it changed, and cron or launchd runs them. Watching a page for changes has the script, and compares one section by its heading so a clock or a rail of new posts elsewhere on the page does not count. Nothing runs in the background between scans, and nothing judges which changes matter.

Can several browser profiles connect?

Yes. The machine has one pairing, and every profile that connects shares it.

What does the helper need?

Node 22 or later where install ran. Chrome starts the helper without your shell's PATH, so run install again after removing that Node or installing another browser. It covers the Chrome Web Store and Edge Add-ons builds and the unpacked builds in your browser profiles; if you loaded one seconds ago, run install again. On Windows it is set up through the registry, and has not been tried on a Windows machine yet.

The Obsidian plugin

Why does the Obsidian plugin need Node.js and npm?

It runs the page-scanner command, and Obsidian cannot run it with its own runtime. Install Node.js 22 or newer and npm install -g @page-scanner/cli once; the plugin's settings show what is missing and check again as you go. Obsidian's community plugins may not install anything for you, so the plugin never runs npm itself.

On this page