Sabi Cap: the cap with 100,000 EEG sensors that promises to turn thoughts into text

The California startup Sabi, based in Palo Alto, has introduced a cap equipped with EEG sensors (electroencephalography) designed to translate brain activity into text. The stated goal is to enable writing without physical or vocal input, simply by "thinking" the words.

The device, named Sabi Cap, integrates up to 100,000 miniaturized sensors, each smaller than a lentil, distributed along the inner surface of the fabric. These sensors detect electrical signals from the brain and send them to a proprietary AI model, called the "Brain Foundation Model," which handles the decoding.

According to the company, the system would be able to achieve a speed of about 30 words per minute, a value comparable to some advanced voice dictation solutions, but without the need to speak. The AI model behind the project is said to have been trained on about 100,000 hours of EEG data collected from a hundred volunteers. The goal was to identify common patterns among different individuals to associate specific brain activities with words or concepts.

This approach aims to overcome one of the main limitations of brain-computer interfaces: the high variability of neural signals among individuals. Sabi claims that its model can generalize these differences, allowing it to operate without lengthy calibration phases.

From a technological standpoint, the system combines EEG sensors, proprietary chips for local processing, and cloud infrastructure for data interpretation. The company also states that signals are encrypted before transmission, with training performed on protected data.

One of the distinguishing elements of Sabi's proposal is the absence of invasive components. Unlike solutions such as Neuralink, which require surgical interventions to implant electrodes in the brain, the cap represents a completely external approach. This aspect could promote greater commercial spread, significantly lowering the access threshold. Furthermore, a non-invasive solution is more easily adoptable in everyday contexts, including work or home environments.

Potential applications also include accessibility: people with motor disabilities could benefit from a system capable of directly converting thoughts into text or digital commands.

Despite the promises, the scientific community maintains a cautious stance. Recent studies highlight how the effectiveness of EEG-text models is still uncertain, partly due to evaluation methodologies that are not always robust. Some researchers suggest that the most promising results obtained so far may stem from phenomena of "pattern memorization," rather than from a true ability to decode the semantic content of brain signals. In other words, the models might recognize recurring patterns without truly understanding the content of the thought.

Another obstacle is the quality of the signal: non-invasive EEGs, operating from outside the skull, offer lower resolution compared to implanted electrodes, making accurate interpretation more complex.

Sabi plans to launch the device commercially by the end of 2026, with a possible variant in the form of a cap with a visor. No prices, availability, or additional features beyond text transcription have been announced yet. In the absence of public demonstrations or independent performance data, it remains to be seen whether the product will be able to deliver on its promises. The risk is that the technology may prove to be less revolutionary than claimed, at least in the initial phase.