Pull on what looks like an ordinary winter hat. Think the words you want to write. Watch them appear on your screen—no typing, no voice commands, no implants drilled into your skull.
That's the pitch from Sabi, a Silicon Valley startup that slipped out of stealth mode in mid-April with a promise audacious even by the standards of an industry fond of moonshots. The company says it will ship a brain-reading beanie by the end of 2026, one capable of translating imagined speech into text at roughly 30 words per minute. For perspective, that's the threshold some invasive brain-computer interfaces have hit in clinical settings—the kind that require neurosurgery and months of calibration.
Sabi wants to do it with a product you can order online and wear to a coffee shop.
Whether the technology can deliver on that timeline remains an open question. But the ambition alone signals how far the brain-computer interface race has come, and how much investor appetite exists for noninvasive alternatives to the surgical implants being pursued by companies like Neuralink.
Reading Thoughts Through Fabric
The technical approach centers on density. Traditional EEG headsets—those research-grade caps bristling with electrodes—typically deploy a few dozen sensors to capture electrical activity from the scalp. Sabi's beanie, according to CEO Rahul Chhabra, will pack between 70,000 and 100,000 miniature sensors into the fabric. That's not a typo: a hundredfold increase in spatial resolution, managed by custom chips the company designed to process the flood of neural data.
When you silently rehearse a sentence in your head—what neuroscientists call "inner speech"—specific patterns ripple across your brain's language centers. Sabi's sensor array picks up those electrical traces, encrypts them (more on that shortly), and beams them wirelessly to the cloud. There, a proprietary AI model attempts to decode the signal and spit back the words you were imagining.
The system runs entirely without wires. A baseball cap version is also in development, Chhabra told WIRED, for users who find beanies less practical in warmer months or professional settings.
Does it actually work at the promised speed? No independent testing has surfaced yet, and the company hasn't published peer-reviewed benchmarks. The 30-word target represents an intention, not a verified outcome.
The Dataset Question
Sabi's core bet is that scale solves the problem. Traditional BCI research often requires lengthy calibration sessions tailored to each user's unique brain anatomy and signal patterns. Sabi claims to have sidestepped that bottleneck by training its model on roughly 100,000 hours of brain data collected from around 100 volunteers—though the claim of having "the world's largest neural decoding dataset" remains unverified by independent audits.
The promise is enticing: a plug-and-play device that works out of the box because the underlying AI has already learned to generalize across different heads, different thinking styles, different neural quirks.
This cross-subject approach mirrors a broader shift in the academic literature. Several papers published in 2025 and early 2026 have explored the idea of "foundation models" for brain signals, borrowing the paradigm that gave us large language models like GPT. But whether Sabi's dataset claim stands up to scrutiny—whether it's truly the largest, or just the largest assembled by a private company—remains impossible to verify without third-party audits. The company hasn't opened its data or methods to outside researchers.
Your Brain, Encrypted (Hopefully)
If reading typed words off a screen feels invasive, consider what reading them directly from someone's brain might reveal. Neural activity doesn't just encode the sentence you're trying to type. In theory, it could leak mood, distraction, even stray thoughts you never intended to share.
Sabi says it's addressed this with end-to-end encryption. Once the beanie captures your brain signals, the data gets encrypted before leaving the device. The company's AI model, Chhabra explained to WIRED, trains on encrypted neural data rather than raw signals.
That raises immediate technical questions. Homomorphic encryption? Secure multi-party computation? Some novel architecture the company developed in-house? The details haven't been disclosed, and encryption schemes that allow computation on encrypted data are notoriously complex and slow. How Sabi squares real-time decoding with robust encryption is unclear.
Also unclear: whether the device will require regulatory approval. Sabi is positioning the beanie as a consumer computing accessory, not a medical device, which could allow it to sidestep FDA oversight. That's a common strategy in the BCI space, but one that tends to attract scrutiny once products reach the market.
Backers and the Noninvasive Bet
The company has raised undisclosed funding from Khosla Ventures, Accel, and Initialized Capital, along with backing from Kevin Weil, a former product executive at Twitter and Instagram. The investor lineup suggests a sizable seed or Series A, though exact amounts weren't disclosed.
Vinod Khosla, in remarks to WIRED, framed noninvasive wearables as the only plausible route to mass adoption. His logic: billions of people might wear a smart beanie. Very few will opt for brain surgery. It's a pointed contrast to the invasive interfaces being developed by Neuralink and Synchron, which target patients with paralysis or neurological conditions but face obvious barriers to consumer scale.
Perhaps more than the founders expected, the race to commercialize BCIs has become a referendum on risk tolerance. Implants offer higher signal fidelity but demand medical procedures. Wearables promise convenience but wrestle with noisy, lower-resolution data. Sabi's 70,000-sensor gambit is an attempt to close that gap.
The Year-End Test
Sabi opened a waitlist in mid-April, though pricing, battery specs, and other practical details remain undisclosed. The launch timeline—products shipping before 2027—is aggressive by any measure, especially for hardware involving custom chips, novel AI, and untested regulatory pathways.
If the beanie ships on schedule and hits its performance targets, Sabi will have pulled off something rare: moving thought-controlled typing from the realm of research labs and paraplegic trials into the hands of ordinary consumers. That would be a genuine inflection point for the industry.
If it doesn't? Well, it wouldn't be the first BCI startup to discover that brains are considerably harder to read than venture pitch decks suggest. The gap between what neural data might reveal and what it reliably does reveal has humbled more than a few well-funded teams.
By December, we'll have an answer.
