OpenAI killing Atlas pushed me to a local AI browser setup, and I should have made the switch much sooner

Published Sep 11, 2026, 7:00 PM EDT Nolen began their writing career in 2019, with three years dedicated to editing the Creative section at MakeUseOf. Their expertise lies at the crossroads of technology and creativity, covering areas like photography, video editing, and graphic design. Outside of work, you'll often find Nolen diving into a good book, writing their own stories, or playing video games. When OpenAI announced in July that ChatGPT Atlas would stop working in August 2026, less than a year after its October 2025 launch, the shutdown was less surprising than the timing. Nine months from launch to sunset for a browser that was supposed to be OpenAI's big play at Chrome, folded back into the ChatGPT desktop app and a Chrome extension instead. Many of the AI tools I've watched come and go over the past two years have followed a similar route. There's usually some big launch, a lot of hype and tutorials and reviews, then like clockwork there's a blog post explaining that the team is being redirected somewhere else. I already run a portion of my AI stack on local models for exactly this reason, though I haven't extended that to my browser yet, until recently. Want to stay in the loop with the latest in AI? The XDA AI Insider newsletter drops weekly with deep dives, tool recommendations, and hands-on coverage you won't find anywhere else on the site. Subscribe by modifying your newsletter preferences! The AI graveyard is getting a little crowded Atlas is not the first, and it won't be the last Atlas itself is the cleanest recent example. OpenAI launched it in October 2025 as a new agentic browser built around ChatGPT, then confirmed the sunset for August 9, 2026, nine months in. The official framing was that the features would live on inside the ChatGPT desktop app and a Chrome extension, though internally OpenAI's leadership had been telling teams to cut what they called "side quests" for months. This is the same reason Sora got shut down back in April, which left mid-project creators unable to access videos they'd already generated. To be fair, Atlas did earn some of its ending, with prompt injection vulnerabilities documented, and one test also clocked the agent at ten full minutes to add three items to an Amazon cart. Arc is another one, and a browser I'd heard designers talk about non-stop a while back. The Browser Company put it into maintenance mode in May 2025 and pivoted the whole team to a new AI-first browser called Dia, and then Atlassian bought the company for $610 million in October. Arc still runs but nobody's building it anymore, and the source code stayed closed because it shares an SDK with Dia. And even the tools that don't fully shut down don't really sit still - OpenAI killed GPT-4o inside ChatGPT this February and a Change.org petition with more than 20,000 signatures didn't bring it back. I don't think any of this is out of malice, it's just incentives that control the situation. Why use local AI in the browser at all? When the browser itself is already online My browser is a cloud thing by definition; it talks to remote servers all day, logs me into accounts, and syncs my history. So what does swapping out the AI layer for a local model actually change? The distinction, I think, is between the browser's normal traffic and the AI layer sitting on top of it. Normal traffic goes to sites I've already decided to trust, or at least tolerate. The AI layer is a different thing because it sees everything at once, including the page I'm reading, the doc I've asked it to summarize, and the prompt itself where I've explained what I'm actually trying to do with it. That's the part I want on my own machine. So when I run a local model through BrowserOS pointed at LM Studio, the contents of every AI conversation just stay on my SSD, my prompts aren't training data for anyone, and if either BrowserOS or LM Studio ever gets shut down, I've still got my past chats. Local isn't the answer to everything though. Ad tracking, ISP visibility, and fingerprinting are all separate problems that come down to which browser you pick in the first place. There's also a capability issue; BrowserOS's own docs say local models handle Chat Mode well but aren't yet powerful enough for Agent Mode, which needs a bigger brain to reliably click the right elements across multi-step tasks. I've been doing just fine without Agent Mode, however. My setup (and the alternatives worth knowing) It took five minutes to set up My setup is pretty boring, with BrowserOS as my browser, LM Studio running a local model as an OpenAI-compatible server, and the browser pointed at that endpoint. If you're interested in switching to BrowserOS, it lets you import Chrome data since it's a Chromium fork, so your bookmarks, passwords, and extensions can carry over smoothly. I prefer BrowserOS specifically because it treats local as a first-class option rather than an afterthought. It also works on Windows, Mac, and Linux, with built-in MCP support for Gmail and Docs and a handful of other useful nuggets. And if the company behind it vanished tomorrow, the code's public and someone could fork it, which is the actual insurance policy against everything I was just complaining about. Brave is the other one I'd recommend for local AI chat. Leo's Bring Your Own Model feature hits any OpenAI-compatible endpoint, so LM Studio and Ollama both plug in smoothly. The only downside is that the "Answer with AI" button in Brave Search itself doesn't route through your local model, only the Leo sidebar does. Then there's also Opera. It was actually the first mainstream browser to ship built-in local LLM support back in April 2024. Opera One now covers over 2000 model variants through the Ollama framework. My AI browser can't be sunset, sold, or gutted in a policy update Watching Atlas vanish is what ultimately pushed me to start bringing local AI to my browser workflows, and also choose more reliable and open browsers. Every AI tool you don't own is basically on a clock, whether it's a browser or a video model. So it might be worth looking into ways you can make pieces of your AI stack more your own. BrowserOS