Silicon Valley's Plan: A Beanie That Reads Your Mind

In Brief

Sabi, a California-based startup, came out of stealth in April 2026 with a product that sounds like science fiction: a winter hat packed with 70,000 to 100,000 tiny EEG sensors, designed to turn silently-imagined words into text on a screen. The company is backed by Khosla Ventures, Accel, Initialized, and former OpenAI chief product officer Kevin Weil.

The pitch, from CEO Rahul Chhabra, is a trade-off. Their type of brain-computer interface, known as BCI, doesn't require surgery. But signals read from outside the skull are weaker than what you get from an implant. Sabi's bet is that if you throw enough sensors at the problem, and train a big enough AI model on the data, you can decode useful thoughts anyway.

The catch: their first public target is 30 words per minute, slower than most people type on a keyboard. And even friendly coverage warns that "reading thoughts" is overstated for what EEG can actually do today.

What Sabi is actually building

The product is a winter beanie with 70,000 to 100,000 miniature EEG sensors woven directly into the fabric. EEG is a technique for measuring the brain's electrical activity from outside the skull. A normal medical EEG cap has a few dozen sensors. Sabi is trying to do the same job with a thousand times more.

To turn those signals into text, Sabi has trained what it calls a brain foundation model, using 100,000 hours of neural data collected from 100 volunteers. A foundation model is a large AI model trained on a wide dataset, so other tools can be built on top of it. Here, the "other tool" is a keyboard you operate by thinking.

Chhabra puts the technical claim this way: "With that high-density sensing, it pinpoints exactly what and where neural activity is happening. We use that information to get much more reliable data to decode what a person is thinking."

The honest trade-off: no surgery, weaker signal

"Invasive" means something is surgically placed inside the body. "Non-invasive" means it just sits on the outside. Here is the comparison Sabi is making:

Think of it this way: reading brain activity through the skull is a bit like listening to a conversation through a wall. You can tell there are voices, and with enough training you can guess what is being said. But you are always a layer of bone and tissue away from the source.

Implants are the microphone in the room. Sabi's beanie is the ear pressed to the wall. The only way to close that gap is software: more sensors, smarter models, and enough training data to compensate for each sensor being individually weaker than an implanted electrode.

Why the pitch works anyway

Vinod Khosla, whose firm is one of Sabi's lead investors, framed the bet bluntly: "The biggest and baddest application of BCI is if you can talk to your computer by thinking about it. If you're going to have a billion people use BCI for access to their computers every day, it can't be invasive."

That is the entire investment thesis in two sentences. The mass-market opportunity requires something you can put on and take off. Brain surgery does not scale to a billion users.

Where the skepticism lives

Coverage sympathetic to Sabi's approach still points out real limits:

• Signals vary between individuals. Two people thinking the same word will produce different EEG patterns. The same person thinking the same word will produce slightly different patterns each time.
• "Mind-reading" is overstated. Current EEG-based systems can decode limited patterns or commands. Continuous, natural thought-to-text is still an evolving challenge, not a solved problem.
• 100 volunteers is a small training set for a system that has to generalize to anyone who puts on the hat.

For context on speed: an average keyboard typist hits 40 to 50 WPM, and a fast one 80 or more. Sabi's initial 30 WPM target is not a promise to beat typing. It is an alternative for moments when your hands are not free — a text message during a meeting, a note during a lecture, a quick reply without picking up the phone.

The privacy question nobody has answered

Brain data is the most personal category of data that exists. A device that continuously reads neural activity, even clumsily, sits closer to your inner world than anything in your phone.

Sabi says it encrypts the data. That is a true answer to a narrow question. The bigger question is what actually gets collected, what gets sent to servers, and what the company promises never to do with it. Those answers have not been published yet, and they are the ones that matter before a product like this ships to a consumer.

The real shift

The interesting thing about Sabi is not the technology. EEG caps that try to do this kind of decoding have existed in research labs for years. The shift is ambition: a consumer product, priced for millions, that you wear like a hat.

Neuralink and its invasive peers have spent a decade proving that BCI can work for patients with severe disabilities. Sabi is a bet that the next decade is about making it work for everyone else.

Whether the beanie ships on time at 30 WPM, or whether a sharper competitor catches up first, the framing has already changed. BCI is no longer only a medical story. It is a consumer-product story now, with all the privacy, speed, and hype questions that come with that.

Glossary

Sources and resources