IBX Labs
A research house.
Cognitive AI models. Non‑transformer architecture.
Totality, as seen from the house.
Intelligence is prediction. The brain does not compute an answer. It settles into one.
The field builds larger and larger memories and calls the recall thinking.
We build models that think in compute, not in tokens — models that work until they converge, and stop when they know.
The transformer ate the field. We are eating the transformer.
ماهر
māhir, noun. One who is skilled; a master of a craft.
Maher is a language model with no attention and no transformer blocks. It is built on predictive coding: columns of units that compete to predict what comes next, and learn from the error when they are wrong. Where a transformer spends the same computation on every token, Maher works until it converges — and stops when it knows.
Adaptive compute
Hard tokens receive more thought. Easy ones, less. The model allocates its own effort.
Native memory
Long‑range memory lives in the substrate, not in a window.
Continuous learning
The model learns as it runs. What it learns, it keeps.
IBX Labs was founded by Ibrahim Maher in Cairo.
Two years of research.
Built alone, so far.
The work goes where it must.
If you are building what comes next, write.
ibrahim@ibxlabs.comOpen to researchers, compute, and correspondence.