Sam Altman posted last month that ChatGPT has over 900 million weekly users. I saw the number and my first thought was: almost all of them are using it in a tab that cannot see the tab next to it. And until last week, dassi required an API key to connect to any model, which meant the vast majority of those 900 million people had no easy path to browser-native AI.

That changed. You can now sign in with your existing ChatGPT Plus or Pro subscription and start using GPT inside Chrome immediately.

Why API keys were the wrong first step

I watched a friend try to set up an OpenAI API key last month. She’s a marketing manager, extremely competent with SaaS tools, runs campaigns across four platforms. She got to the API billing page, saw that it was separate from her ChatGPT subscription, and closed the tab. “I already pay twenty dollars a month for this,” she said, and that was that.

She’s not wrong. The API key workflow assumes you understand the difference between a subscription product and a platform API, that you’re comfortable with usage-based billing, and that you know what a “secret key” is and why you should not paste it into random websites. Most knowledge workers, the people who would benefit most from browser-native AI, bounce at step one because the entire onboarding path was designed for developers.

Signing in takes about twelve seconds

Open dassi’s side panel in Chrome. Click the ChatGPT login option. Sign in with your existing account. That is the entire process, and I timed it because I was curious whether the OAuth flow would add friction, but it turns out redirecting to a login page you’ve already used a thousand times is the opposite of friction.

The model immediately has access to whatever page you’re looking at. No pasting, no explaining context, no switching tabs. You’re reading a long product comparison on some review site and you ask “which of these has the best return policy” and the answer comes back referencing specific cells in the comparison table you’re staring at. That interaction is impossible on chatgpt.com because chatgpt.com does not know the page exists.

The people who actually need this

I keep thinking about who benefits from removing the API key barrier, and it is not developers. Developers were fine, they had API keys already, many of them prefer the granular billing control. The people this unlocks are:

Recruiters screening LinkedIn profiles who want to draft outreach without copy-pasting every candidate’s experience section into a separate window. Account managers who live in Salesforce or HubSpot and need to summarize client histories that span dozens of activity entries. Researchers reading dense PDFs in the browser who want to ask questions about the content without re-uploading the document somewhere else.

These are people with ChatGPT subscriptions who use the product daily but have never heard of an API key, and frankly should not need to hear about one just to get AI working where they actually work.

What about power users

API key mode is not going anywhere. If you want Claude Opus for careful analysis, Gemini 3.1 Pro for its million-token context window, or DeepSeek for cost efficiency, you still bring your own key and switch models whenever you want. The subscription login is a second door into the same product, not a replacement for the first one.

And honestly the two modes serve different philosophies about how people relate to AI tools. Some users want maximum control and model selection. Others just want the damn thing to work when they click it. Both are valid and both are now supported.

Browser context is the whole point

I want to be specific about what “works inside your browser” means because it gets hand-wavy fast. When you ask GPT a question through dassi, the model receives the DOM of the page you are currently viewing. Not a screenshot, not a URL that it then fetches independently from some cloud server. The actual content of your authenticated, logged-in, real browser session. So when you’re looking at an internal dashboard behind your company’s SSO, or a Gmail thread, or a Notion doc that requires login, GPT can read it because your browser already has access.

This is different from cloud-based AI agents that spin up a remote browser instance and cannot see anything behind authentication. Your local Chrome is already logged into everything. That is a massive advantage that gets undersold, partly because “we read the DOM” sounds less exciting than “autonomous AI agent,” but it is the technical detail that makes the whole thing actually useful for real work rather than demo-worthy toy tasks.

Getting started

Install dassi from the Chrome Web Store. Sign in with ChatGPT. Open a page you were going to copy-paste from anyway. Ask a question about it instead.

It’s a small change in workflow that eliminates a surprisingly persistent source of daily friction, the kind of friction that most people have stopped noticing because they have been doing it since 2023.