Tomás Vega went to Neuralink, the Elon Musk-founded startup, to work on the highest-bandwidth human-machine interface possible to liberate people from their disabilities. Then he left because he couldn’t wait for it.
Vega, who has stuttered since he was a kid, believes implantable brain-computer interfaces and artificial intelligence will change the lives of disabled people in the future. But he wanted something that could help now—without surgery. “I couldn’t let my friends who live with quadriplegia wait so long,” he says. So Vega left Neuralink to return to graduate school at MIT’s Media Lab. He wanted to build an interface “as expressive as an implant without all those constraints.”
So, alongside cofounder Corten Singer, he created MouthPad with Augmental.
MouthPad looks like half a retainer that is placed against your palate and back molars. At its center there’s a golden capacitive trackpad. A tongue press applies positive pressure and produces a left-click; a sip applies negative pressure and produces a right-click. In tongue-control mode, touch location becomes a virtual joystick. In head-control mode—which works using its internal gyroscopes if the user can move their head—head movements move the cursor while tongue touch operates virtual buttons for vertical and horizontal scrolling.
The device doesn’t seek to reproduce Musk’s brain-interface ambition more cheaply or more discreetly. It rejects the premise that high-agency human-computer interaction must begin with surgery—or that accessibility must announce itself on the wearer’s face with some cyborg machine interface.
MouthPad seeks to use the body’s remaining abilities. Its most provocative proposition is that the best assistive technology might be the kind that gives someone control without forcing them to look like they are using assistive technology. Basically, it’s a mouse that disappears into your mouth. “We [Augmental] are people-first,” Vega says. “Technology should disappear so that you see a human involved.”
The long way to the tongue
Vega’s time at Neuralink clarified his approach, but it did not create it. The roots of the idea go back to when he was five and began stuttering. He says a keyboard and mouse helped him work around limitations he experienced in speech. “I became obsessed with input devices,” he says.
Vega started programming at 11, and worked with people with multiple sclerosis in high school before studying brains and computers at UC Berkeley. He tried scalp electrodes, EOG, EMG, and voice interfaces. He even shaved his head daily for an electrode-based setup, he says. They could make for arresting demos, Vega tells me with a smile, but they were not useful enough in everyday life.
“But the better I understood how the brain and the body work, the better I could probe it and the better I could hack it,” he remembers. “And the better interfaces I could design to augment human ability.”
After leaving Neuralink, Vega began with a more practical question: What part of the human body still has extraordinary range, precision, and endurance for people who cannot use their hands? “Thinking from first principles, what is the area of the body that is the most expressive?” he recalls. “That’s our tongue.”
The first prototype was a sensor-packed lollipop called “Chupetín” (“small pacifier” in Spanish). Its purpose was to test the user experience of a tongue-computer interface. “It worked great,” Vega says. “It was intuitive. I didn’t have to think about it.” That led to Augmental, which Vega says started in a basement in Somerville, Massachusetts.
A mouse built like a retainer
MouthPad is custom-fit through an intraoral scan of the user’s mouth. That matters because the device needs to sit thinly on the palate—the part the tongue touches—while still allowing the user to speak. Augmental began with putty impressions but moved to optical intraoral scanning after finding too much human error in physical impressions.
Building the hardware was an exercise in hostile-environment design. The device uses a serpentine microfabrication approach applied to flexible electronics, so one electronic design can stretch and compress to accommodate different mouth shapes and sizes. A bridge passes through the occlusal plane, where the teeth meet, creating a protected channel so a bite does not destroy the circuitry.
The earliest version used pressure-sensing films. Those deformed and affected the baseline signal, Vega says. The company now uses capacitive sensing—the same fundamental technique used in phones and laptop trackpads—to detect the tongue’s position on the palate. A small air chamber and barometer register pressure changes, allowing MouthPad to distinguish a tongue press from a sip.
That is the physical interface. The harder part is making it behave like an interface rather than a science experiment. That’s where software comes in. MouthPad’s algorithm analyzes a matrix of sensor data to infer intent and identify tongue position despite saliva. Vega compares the problem to using a touchscreen on a rainy day.
The device is meant to connect as a computer mouse. Users can point, left-click, right-click, and scroll through tongue gestures; those who can use head control can move the cursor with head movement and use the tongue for other functions.
Beyond point and click
Computer mice are limited tools, which is why the modern computer is reliant on keyboard shortcuts, contextual commands, tiny click targets, and repetitive actions that would otherwise exhaust anyone forced to do everything through a pointer.
Augmental’s answer is a companion app that layers profiles and gesture-triggered shortcuts on top of the core mouse interface. Its central interaction is a context radial menu that can be opened by remapping the sip gesture. Once it appears, the user can select on-screen virtual buttons with tongue touch location.
That can make everyday operations less punishing. The radial menu can offer common shortcuts such as copy and paste without forcing someone to aim at a tiny right-click menu. Vega says Augmental has also demonstrated functions for dictation and AI-agent commands.
The idea is not that the tongue becomes a keyboard in the familiar sense. It is that profiles can make the same gesture mean different things in different contexts. In a first-person-shooter profile, head movement controls aim via the mouse while tongue touch generates WASD movement. In a mechanical CAD profile, the context menu can trigger keyboard shortcuts for pan, orbit, and zoom, while head movements control the mouse for the selected translation or rotation.
Augmental calls this a five-degree-of-freedom computer controller: two degrees of freedom from head-pointer movement, and three from tongue gestures. The point is not a marketing-friendly number. It is the company’s effort to recover some of the expressive range that hands bring to computing through continuous control, pressure, sequences, and context.
The performance argument
Augmental’s report—which is company-authored and not peer reviewed yet—says it has worked with approximately 150 people during nearly two years of early access. Its point-and-click dataset includes 58 people with hand impairments who completed a one-minute grid-selection game and recorded a score; the group included people with spinal-cord injuries, quadriplegia likely related to SCI but unstated, ALS, multiple sclerosis, muscular dystrophy, repetitive stress injury, and other conditions.
In the test, 44 people with limited hand use controlled a computer cursor by moving their heads and using MouthPad with their tongues. The company says their typical score was 3.43 bits per second (BPS)—a technical measure of how quickly and accurately someone can select targets on a screen. Thirty-four people tried the MouthPad test on their first day using it. The middle score was 3 BPS, and five people scored above 5 BPS. Augmental says users generally learned the basics in less than two hours.
But the numbers have limits. The test asked people to click targets on a screen arranged in a 30-by-30 grid. It did not test how fast they could type, whether the device caused fatigue, whether they could use it independently, or whether it helped them work through a normal day. Augmental says its fastest MouthPad user outperformed the fastest users of the other systems included in its comparison. It also says that 45.45% of its head-control users scored higher than the best result in the published brain-computer-interface study it cited. The company did not include Neuralink’s public results in that comparison because those results have not been reviewed by outside scientists.
This is not a case of a mouth retainer “beating” an implanted brain-computer interface. It is evidence that a noninvasive system can provide meaningful computer control for people who retain tongue control and, in head-control mode, head movement.
The company’s own review acknowledges that camera-based head tracking and eye tracking can produce better typical performance than MouthPad in the Fitts’ law comparisons it cites for typically abled participants. But head control is not physically possible for every user, can be uncomfortable while lying on one’s back, and can become noisy in a bumpy vehicle. Plus, MouthPad is invisible.
The case for invisibility
For Vega, the point of placing the hardware inside the mouth is not merely elegance. It is social. “Because our users, many of them use interfaces that are on their face, like the joystick and whatnot,” he says. “That creates a lot of stigma.” MouthPad is an assistive interface that does not sit between its wearer and everyone else in the room. This does not make visible assistive technology lesser, nor does every user want their technology hidden. But invisibility can be an important form of control—one more decision that belongs to the person wearing the device.
The mouth is also useful because it leaves speech intact. That was a design requirement from the beginning, Vega says. Speech remains one of the fastest ways to produce text for people who can use it, but ordinary voice control can be loud, public, and unreliable in noisy places.
That is why Augmental has built Vox, its next product. Vega describes it as a “stethoscope-like microphone” optimized for voice traveling into the body. “It listens to you, not to the world,” he says. Vox originated as a low-volume speech feature for MouthPad users who may have reduced lung function, impaired diaphragm function, tracheostomies, altered phonation, or no privacy when using standard voice interfaces. The company describes it as a microphone engineered to capture low-volume, body-conducted speech in noisy environments.
Together, the two products are supposed to close the loop: MouthPad can explicitly start and stop dictation; Vox provides the quiet voice signal. Augmental calls the pairing an efficient mouse-and-keyboard replacement, though it has not published a formal performance study of the combined workflow.
For Vega, that use case is personal too: “When I whisper, I don’t stutter as much,” he says. He also says Vox has allowed him to work while walking, doubling his average daily step count from 8,000 to 16,000. “I’ve reclaimed an hour under the sun,” he says, “and I think that’s beautiful.”
He says Augmental has shipped 180 devices and has more than 120 users. Since the device’s battery lasts about seven hours, some users have bought second or third devices to cover longer days or keep a backup.
While the product has not been approved by the FDA yet, the company works with rehabilitation centers and occupational therapists in the United States. MouthPad costs $1,400, according to Vega, and once they get the FDA’s green light, insurance companies will reimburse part of that cost. Vox costs $200. Vega wants to expand to Europe, Japan, and other markets while lowering MouthPad’s price toward mainstream consumer-electronics territory.
He says that eventually, both devices could evolve to jump into the mainstream market. Vega does not want to rush that move, he tells me as he points to Humane and Rabbit’s failures as warnings against scaling hardware before it is ready: “Let’s go fast. Let’s iterate. Let’s have new versions every quarter. Let’s learn. And when the market screams for it, then we activate it.”
That may be the most revealing difference between MouthPad and the usual brain-interface sci-fi fantasy that’s still years away from materializing in the real world. The fantasy begins with a superhuman future and asks people to wait for it. Vega’s project begins with the person who needs an interface today. And then we can see where that leads us.
MouthPad’s version does not begin with drilling into a skull. It begins with an ability many people still possess, a tongue moving against the roof of the mouth or a hidden microphone that listens to your whisper through your flesh and bones. Devices small enough to vanish and useful enough to make big problems disappear.