If you’ve ever thought about Harry Potter and wished for a sorting hat that could read the thoughts and feelings of new students at Hogwarts School of Witchcraft and Wizardry, your wish has now come true. Palo Alto-based startup Sabi has emerged from stealth mode and unveiled a hat that can read minds.

This is supposed to work via a brain-computer interface (BCI). So far, we’ve seen approaches like Elon Musk’s Neuralink, where a hole is drilled into the skull and a chip the size of a button battery is implanted and connected to the brain cells via wires.

Sam Altman, founder and CEO of OpenAI—known for ChatGPT—is also interested in BCIs. For instance, he has invested in the startup Merge Labs, which aims to read thoughts using ultrasound.

And Sabi’s BCI is said to work in exactly the same non-invasive way—that is, without the need for surgery, as is the case with Neuralink. The technology for reading thoughts is based on electroencephalography (EEG), in which metal electrodes are attached to the scalp to record the brain’s electrical activity. Decoding thought-based language using EEG is already possible, but is currently limited to short phrases or commands and does not yet support continuous, natural speech.

The drawback of a non-invasive BCI is that the sensors must detect brain signals through a layer of skin and bone. This attenuates the neural signals. Surgically implanted devices receive much stronger signals because they are located so close to the neurons. Sabi aims to improve the accuracy of a non-invasive device by massively increasing the number of sensors. Most EEG devices have between a dozen and a few hundred sensors. Sabi’s cap will have between 70,000 and 100,000 miniature sensors.

This headset is designed to help users achieve an initial typing speed of about 30 words per minute. While that’s slower than most people type, the speed is expected to improve the more time users spend with the headset.

To enable thoughts to be deciphered at all, Sabi developed a large-scale AI model known as the “Brain Foundation Model.” This model is trained using extensive neural data from many people to learn fundamental activity patterns that correlate with inner speech. According to CEO Raul Chhabra, the company has collected 100,000 hours of brain data from 100 volunteers to date. The data is encrypted throughout the entire process as it leaves the device and is uploaded to the cloud.

The beanie is expected to go on sale in late 2026.