TL;DR
OpenAI is shutting down the Atlas browser. It stops working on August 9, 2026. The announcement landed July 9–10, 2026, roughly nine months after launch. OpenAI's stated rationale: "the browser is a feature, not the destination."
If you have anything you care about living in Atlas — research you gathered, threads you saved, pages you meant to come back to — you have weeks, not months. The move that survives the deadline is boring and reliable: open the pages that matter in a browser you control and save them as PDFs, now, while they still render.
That's what Convert: Web to PDF is for. It converts the page in front of you into a real PDF — selectable text, clickable links, embedded fonts — locally, through Chrome's own print engine, including on pages behind a login. It doesn't need Atlas to exist, and it won't care on August 10.
Then the wider point, which we're going to make without gloating, because a lot of people put real work into Atlas and there's nothing funny about a shutdown: a browser is a big thing to depend on. The AI-browser category just demonstrated that the whole floor can be removed on about thirty days' notice.
What's happening, precisely
The facts, kept narrow on purpose:
- July 9–10, 2026 — OpenAI announced it is shutting down the Atlas browser.
- August 9, 2026 — Atlas stops working.
- The browser lasted roughly nine months from launch to shutdown announcement.
- OpenAI's framing: "the browser is a feature, not the destination."
That last line is worth sitting with, because it isn't a retreat. It's a strategy statement, and the rest of the announcement backs it up.
Where Atlas's functions are actually going
This is the part that gets lost in "OpenAI kills browser" headlines. OpenAI isn't exiting the browser. It's unbundling it into three pieces:
- A ChatGPT Chrome extension — page context, summarize, ask, and the ability to launch longer tasks. It lands directly opposite Google's Gemini side panel.
- An upgraded ChatGPT desktop app with a built-in browser — handling login, downloads, and page interaction.
- A server-side cloud browser for agent tasks — the agent gets a browser that runs on OpenAI's infrastructure rather than yours.
Read plainly: OpenAI decided that owning the whole browser wasn't worth it when it could sit inside the browser you already use, plus run its own browser on the server side for the agent work. That's a defensible call. It's arguably the correct call. It just happens to be a call that leaves anyone who standardized on Atlas with an August 9 problem.
And here's the thing that should give you pause even if you never touched Atlas: the functionality survived. The container didn't. OpenAI's AI-browser ambitions are, by their own account, still growing. The specific application you installed is what evaporated. Those are different things, and only one of them was holding your stuff.
Do this first: the deadline checklist
Before analysis, action. You have until August 9. Here's the honest scope of what we can and can't help with, stated up front:
We save pages you can open in a browser. If you can load it in Chrome and see it, you can convert it to a PDF. What we cannot do is reach into an application's internal data store and export it for you — we're not an Atlas migration tool, we don't have one, and nobody should pretend otherwise. For anything that lives only inside Atlas as app data rather than as a page, check whether OpenAI is providing an export path. That's their side of the fence.
With that boundary clear, here's what's worth converting before the ninth.
1. Anything you'd have to pay to see again
Paywalled articles you have legitimate access to right now through a subscription. If you cancel that subscription in October, the page is gone to you. A PDF saved in July isn't.
2. Anything behind a login you might lose
This is the highest-value category and the one people always underrate. Work dashboards. Internal wikis. Portals for accounts you're going to close. Anything where access depends on a session, a role, or an employer.
Convert: Web to PDF uses your already-authenticated session, which means login-protected pages convert like any other page — Gmail, banking dashboards, internal tools. There's no "paste your credentials here" step and there is no login bypass, because there's nothing to bypass: you're already in. We wrote this up in detail in saving webpages behind logins as PDF.
3. Research trails and source pages
If you were using Atlas for research, the sources are the durable asset. The synthesis you can rebuild. The specific page that said the specific thing, on the specific date, is the part that quietly rots — sites redesign, articles get edited, links die.
Convert the sources. Our PDFs are real PDFs, so the text stays selectable and searchable and the links stay clickable — which matters a lot when you're feeding them to something else later, or just want to Ctrl+F your own archive.
4. AI conversation threads worth keeping
If a chat thread contains reasoning you'd hate to lose, it's a page — so it converts. Scroll it fully first (more on why below), then save. We have a whole guide on saving AI chatbot conversations as PDF.
5. Receipts, confirmations, records
Boring, and the category people regret most. Order confirmations, service records, anything you might need to prove happened.
One honest caveat we repeat every time: where a platform offers its own official PDF — bank statements, government certificates, tax documents, tickets with QR codes — go get theirs. That's the authoritative document. Ours is a dated snapshot for your records, and it's genuinely useful for everything the platform doesn't hand you, which turns out to be most things.
6. The stuff you'd only miss in six months
Bookmarks are not a backup. A bookmark is a pointer to something someone else controls, and the entire lesson of August 9 is what happens when someone else's decision arrives.
How to actually do it, briefly
Open the page in Chrome (or Edge, Brave, Arc, Opera, Vivaldi — Chromium browsers; not Firefox or Safari). Then:
- Scroll to the bottom first if the page lazy-loads or infinitely scrolls. We capture what's already loaded in the tab. We don't scroll for you and won't pretend to. Single items beat feeds, every time.
- Turn on Load All Images to force lazy-loaded images to render before capture.
- Use Remove Elements to click away cookie banners, sticky headers, or a chat widget before export. It has undo, so you can be reckless.
- Article Mode strips a page to its main content via the Readability algorithm. One caveat, and it's a real one: it can drop
<pre>code blocks. For code-heavy pages, use the default mode. - Single Page Mode avoids arbitrary page breaks landing in the middle of a table or diagram.
- Keyboard shortcut is Ctrl+Shift+P if you're doing a batch.
Free, no watermark, no account, no file size limits.
The wider point: a browser is a big thing to depend on
Now the part that outlives the deadline.
Software fails all the time. What matters is how much comes down with it. That's a property of what you depended on, not of how good the software was — and Atlas, by all accounts, was not bad software. It got shut down anyway, for strategy reasons, with about thirty days' notice.
So the useful question isn't "was it good?" It's "what's the blast radius when it goes?"
Blast radius, by dependency type
| You depend on | If it shuts down | What's still yours |
|---|---|---|
| A whole browser | Everything routed through it goes at once — sessions, saved state, workflow, the container itself | Whatever you exported before the date |
| A cloud service | Your library sits on their servers until they let you export it, on their timeline | Whatever their export tool covers |
| An account-gated app | Locked to their auth. Your access ends when their business decision lands | Whatever you pulled out first |
| A local single-purpose extension | The tool stops. The files it made are already on your disk | All of it — PDFs don't phone home |
The bottom row is the whole argument, and it's small on purpose. An extension that does one thing locally cannot take your library down with it, because it never had your library. Convert: Web to PDF writes a PDF to your downloads folder and then stops existing as far as that file is concerned. There is no account holding it. There's no server that has to stay up. If we vanished tomorrow, every PDF the extension ever made would sit exactly where it is, opening in any PDF reader on earth, in twenty years.
That's not us being clever. It's just what happens when a tool doesn't own your data — the failure mode gets bounded by the tool's own smallness.
The dependency question isn't about trust
Here's what we want to be careful about, because it would be easy to turn this into a cheap "see, we're trustworthy and they're not" pitch. That's not it. OpenAI didn't betray anyone. They shipped a browser, learned the browser was a feature rather than a destination, and acted on it. That's a company making a rational product decision.
The point is that your exposure shouldn't depend on other people's product decisions being convenient for you. No amount of trusting OpenAI would have kept Atlas alive. The trust was never the failure point. The dependency shape was.
Which is why the interesting axis isn't "good company vs. bad company." It's "how much of my stuff is standing on this?" A big vertically-integrated container is a great experience right up until it isn't yours anymore. A tool that hands you a file and forgets about it can't do that to you — not because it's virtuous, but because it structurally can't.
What we'd owe you if this happened to us
We should answer the obvious return question. What happens if we shut down?
Two things. First, whatever the extension already made is already yours — that's the entire point of local-first, and it's not a promise so much as a description of where the files physically are. Second, our manifesto has this in writing: if we shut down, we publish the source and walk. We will not be acquired into adware.
You should weigh that exactly as much as you weigh any promise from a small shop on the internet, which is to say: less than you weigh the fact that the PDFs are already on your disk and don't need us. That's the part that's structural, and structural beats sincere every time.
Where the AI browsers were genuinely better
Being fair costs nothing and the alternative is being wrong.
There are real workflows where an agentic browser beat a converter and still does. If you want a synthesis across ten sources rather than ten separate PDFs, an AI browser is the right tool and we're not. If you want an agent to read four tabs and produce a writeup, that's not what a print engine does. We've compared the categories honestly before in Atlas vs Comet vs Chrome for saving webpages, and covered converting agentic browser outputs to PDF.
The distinction we'd actually defend: AI browsers are good at understanding pages. They are a risky place to keep them. Understanding is a service you rent. Keeping is a thing you should own. Atlas's shutdown didn't prove the first half wrong — it proved the second half right.
And notice that the useful half survived. The summarizing, the asking, the page context — that's moving into a ChatGPT Chrome extension. The part that died was the container you kept things in. If your setup had kept the understanding rented and the keeping local, August 9 would be an inconvenience instead of a scramble.
Frequently asked questions
What exactly is the Atlas deadline?
Atlas stops working August 9, 2026. OpenAI announced the shutdown on July 9–10, 2026, about nine months after the browser launched.
Is OpenAI getting out of browsers entirely?
No — the opposite, by their own framing. Atlas's functions are moving into a ChatGPT Chrome extension, an upgraded ChatGPT desktop app with a built-in browser (login, downloads, page interaction), and a server-side cloud browser for agent tasks. Their stated rationale is that "the browser is a feature, not the destination." The ambitions grew. The standalone app is what's ending.
Can Convert: Web to PDF export my Atlas data?
No, and we won't imply otherwise. We save pages you can open in a browser. If you can load it in Chrome and see it, we can convert it into a real PDF. Application-internal data stores aren't something we reach into — for that, check whether OpenAI provides an export path.
Do I need to be logged in for the pages I'm saving?
You need to already be logged in, in your own browser, the same as if you were reading the page. We use the session you already have. There is no login bypass and no paywall bypass — those aren't features we have, and we'd rather say so than let you find out at the wrong moment.
Will my PDFs still work if Convert: Web to PDF disappears?
Yes. That's the structural argument, not a marketing line. The PDF is written to your downloads folder by Chrome's own print engine and has no dependency on us afterwards. No account, no server, no phone-home. If we shut down, our manifesto commits us to publishing the source and walking away.
What about long pages and infinite scroll before the deadline?
Scroll first. We capture what's already loaded in the tab — we don't auto-scroll and we're not going to claim we do. For feeds, save individual items rather than the feed. There's a full walkthrough at saving a full scrolling page as PDF and a broader take in the best way to archive webpages.
Is a saved PDF an official document?
No. It's a dated snapshot for your records, and that's a genuinely useful thing to have. But where the platform offers its own official PDF — bank statements, tax documents, government certificates, tickets with QR codes — get theirs. Theirs is authoritative. Ours is for the other 95% of the web that never offers you one.
Which browsers does this work in?
Chromium: Chrome, Edge, Brave, Arc, Opera, Vivaldi. Not Firefox, not Safari. More answers in the 117-question FAQ.
Bottom line
Atlas stops working August 9, 2026. If anything you care about is in there, this month is when you deal with it, and the deal is straightforward: open the pages in a browser you control, and save the ones that matter as PDFs while they still render.
The bigger lesson costs nothing to learn from someone else's shutdown. OpenAI made a rational decision — the browser was a feature, not the destination — and a rational decision by someone else is still a deadline for you. That's what depending on a container gets you, no matter how good the container was or how well-intentioned the people running it are.
Convert: Web to PDF is the opposite bet, and a deliberately unambitious one. It converts the page in front of you into a real PDF — selectable text, clickable links, embedded fonts, no watermark — locally, through Chrome's own print engine, with zero network requests during conversion. It works on login-protected pages using the session you already have. It's free, with no account and no file limits.
It will never be your destination. It hands you a file and gets out of the way, and that file doesn't need us, or OpenAI, or anyone else to still be around on August 10.
Small software. It can't take much down with it. That was always the point, and this month it's the useful one.