Sabi raises $50M for a cap meant to turn thoughts into AI prompts

Sabi is developing a noninvasive brain-computer interface that fits inside ordinary headwear, aimed at people who want to direct AI without typing or speaking. The startup said it raised a $50 million seed round led by Khosla Ventures, with Accel, Initialized Capital, DST Global, Collaborative Fund, Ascend and Kevin Weil also participating.

In Sabi’s design, EEG sensors read electrical activity through a wearer’s hair without gel, surgery or direct scalp contact. Its model would translate deliberately imagined words into text, convert trained mental commands into prompts for an AI assistant, or predict a likely on-screen action for the wearer to confirm. The cap would link wirelessly to a phone or computer.

Today, that is still a development program. WIRED reported in April that imagined-speech decoding from noninvasive EEG was limited to small sets of words or commands, partly because skin and bone weaken the signal. Sabi’s proposed answer is scale: 70,000 to 100,000 miniature sensors, compared with the dozen to few hundred found in most EEG devices, according to WIRED.

The company says its brain model has been trained on more than 100,000 hours of labeled noninvasive recordings collected as participants imagined words or completed mental tasks. It plans to use the financing to expand that dataset, finish its custom EEG chip, manufacture early devices and add neuroscience, hardware and machine-learning staff.

Putting neural input into an everyday accessory also shifts privacy from a laboratory concern to a product requirement. Sabi says it encrypts data at the sensor and is designing inference to run while that data remains encrypted. Planned controls would let wearers decide when the cap records, which applications receive commands and which actions need confirmation.

Sabi is preparing units for members of an early-access waitlist, but it hasn’t disclosed a price or paying customers. It plans to demonstrate early interactions at CES 2027. The engineering hurdle is concrete: WIRED cited an independent neurotechnology consultant who said consumer brain interfaces must work consistently out of the box rather than demand repeated calibration.