Company
The next interface shouldn't be worn like an event.
We are building one device and one runtime, so that nothing about using a computer has to be demonstrated before it can be done.
Thesis
Every input device so far has asked you to interrupt yourself.
A keyboard asks you to stop thinking about the sentence and start thinking about your hands. A voice assistant asks you to stop what you are doing and announce your business to a room. A gesture system asks you to remember which of nine shapes means "undo".
All three put a layer between intention and the machine. Lumina's entire reason to exist is to delete that layer: to take the sentence once it has formed and carry it the rest of the way, in whatever software you happen to be standing in.
The signal
Subvocalised speech leaves a measurable myoelectric footprint. That is physiology, not speculation, and it is the physical foundation the whole product rests on.
The bottleneck isn't sensing
Getting a signal was solved years ago. The unsolved problem is deciding what the user actually wanted, in the specific application they were looking at, and doing it without asking twice.
Both layers or neither
A band without a runtime is a novelty. A runtime without a band is another app. Owning both is what makes the combination behave like a new capability.
Small device, boring interfaces, no magic claims.
On-device or nothing
Neural signal is the most personal data that exists. Any architecture that requires uploading it has already made an unforgivable trade, so the decoder runs on the band and the useful result travels instead of the raw signal.
Native interfaces over integrations
A product that needs a partnership before it works for you is a product that works for the companies with partnerships. Recruiters, editors, terminals, and browsers do not need to be convinced, because Lumina drives the inputs they already accept.
Published limits
Latency, battery, and accuracy numbers get stated in public, including when they get worse. A device that acts on your behalf has to be auditable, and an auditable device publishes its own spec sheet.
Accessibility first, not adapted
The people with the strongest claim on this technology are the people with the fewest alternatives. Designing for the most demanding use first is also the fastest route to a product that is good for the least demanding one.
Two generations, one decision left.
The band exists and decodes. What we are working on now is not whether it works, but how much of a computer it should be allowed to be.
- Generation 1 · complete Bench prototype. Eight-channel dry sEMG on a rigid PCB, decoding phoneme groups offline. Purpose was to prove the physiology, and it did.
- Generation 2 · in progress Wearable band. Flexible substrate, on-device quantised decoder, sub-GHz radio, inductive charging. Target: 18 g and under 40 ms muscle-to-action, worn daily without thinking about it.
- Next The runtime's permission model. What a device carrying your intent may do by default, what it must ask about, and how those boundaries are set by the wearer rather than by us. This is the decision that determines whether the product is trusted.
- Then Closed pilot. A small group with real workloads and the willingness to report every failure. Sizes and dates are set here, not in a launch plan.
Three ways in.
-
Pilot with your team
A lab, a clinic, a workshop, a studio: anywhere people work with their hands committed and a screen waiting. We are looking for environments where the failure modes are honest and frequent.
-
Build alongside us
Electrical engineers, ML researchers on biosignal decoding, embedded systems, and accessibility specialists. If your work already touches any of those, we would like to read it. Send work.
-
Research access
If you study myoelectric interfaces, assistive technology, or human-computer input, we would rather share data and hardware than guard it. Say what you are working on.
Tell us what you would do with it.
We are running a closed pilot, so the applications we take seriously are the ones with a specific answer to that question. One short paragraph is enough.
Direct
Press
Careers
Four open roles across embedded systems, biosignal ML, and product design. Write to us with work, not a cover letter.