Thinking should be readable, and then improvable.
ALLeX is a research lab. We build a non-invasive brain interface and, on top of it, a model that understands the underlying patterns of a person’s thinking and improves them through interaction.
Almost everything built today makes it easier to produce: more text, more decisions, more output per hour. Almost nothing helps a person understand how they think, and nothing hands that understanding back in a form they can act on. We think that is the most important gap left.
The brain gives off signals all the time. Some of them can be read from outside, without surgery, in a device that sits on the head like a pair of headphones. Read well, they say how fast and how deeply someone is thinking, where attention is, what carries a feeling. That is not the content of a thought. It is the state a mind is in while it thinks. It turns out to be the part that has been missing.
What we are building
A wearable that reads state through the intact skull, and a model that puts that state to use. We build it in two stages, on purpose.
The first stage only reads. The person does nothing extra: no calibration ritual, no setup. The device places the state stream alongside an ordinary conversation with a language model, which learns to read it as context, so that answers arrive in the shape the person can take in right now. Where the signal is poor, the device says “don’t know” rather than inventing a value. It does not talk back.
The second stage builds a profile on what the first stage measured: the patterns a person actually optimises for, the frames they think in, what a task costs them and what it is worth, in their own units. The model then works with the person rather than for them. It says what the person can compute well, what a task really demands, and where value has been assigned badly. Only here does the system begin to speak back, and only where the person has agreed to it in advance.
What we believe
Read before you write. Stimulation is where most brain-interface stories begin. It is the wrong place to start. A system that changes how someone thinks must first show, in numbers, that it understands how they think.
Non-invasive or not at all. A technology that helps only those willing to undergo surgery does not reach people. The physics through the skull is hard. That is the work, not a reason to go around it.
Measured, published, set, or open. Every claim we make carries the standing of its source. We say what we have measured, what others have shown under other conditions, what we have set ourselves, and what is still open. We would rather show a sketch than a certainty we do not have.
Comparable across people. A state that cannot be compared between sessions, situations and people is a mood ring. Everything we read is calibrated onto a physical level, so that it stays comparable.
Three brakes. Agreed goals instead of engagement. Costs stated in the unit of the benefit. Full reversibility. Every action the system takes carries its own effect measurement, and the person is always the unit of analysis.
The person owns the profile. What we learn about a mind belongs to that mind. The model exists to hand capability back, not to hold it.
Why now
The building blocks are in print. Signals of state have been seen through the intact skull. Self-calibration of the skull has been shown in people. Wearable ultrasound has run for hours on a head. Language models have become good enough to make use of a few extra words of context. What no one has done is put the blocks together into a device and a model that a person can use every day. That is the whole we are after.
What comes out of it
A by-product of reading many minds well is a description of how humans compute: how much they hold at once, how deep they go, how fast a loop runs, how long attention holds. Once those limits are numbers, they can be checked against an intervention, and where they give, the gain is read off the same number that bounded it before. Models today have outcome data and almost no process data about how a decision came about. This platform produces exactly that.
How we work
We are small and we hire for method rather than for a field: physicists who learn the brain, signal people who learn the skull, model people who learn the person. The combinations we need do not exist as job titles yet. They are made, not hired. Open roles.