Carry knowledge forward.
Construct useful context around a task while keeping a path back to its sources and changes.
Loading ATMAI
About Atmai
Curiosity, made concrete.
Atmai is a research lab working across intelligence, mathematics, and computation. We pursue fundamental questions, build systems to investigate them, and turn selected work into useful products.
Research is the starting point. Building is how we move it forward.
Some questions need a new algorithm. Others need an experiment, a new representation, or a tool that lets someone see the problem differently. We work across these forms because the question should shape the method.
Our programmes investigate continuity of knowledge, coordination between people and agents, evolving computational substrates, and ways to construct and explore mathematics. They share an interest in making the structure of intelligence more understandable.
Products are one outcome of this work. They bring an idea into contact with an actual task, where its usefulness and limits become clearer. Those encounters can open the next research question.
Our agenda
Four ways into a larger question.
What would help intelligence become more continuous, more understandable, and more open to human participation?
Make complex work legible.
Explore visual structures through which people can compose, observe, and direct agent work.
Investigate how a substrate learns.
Study systems in which memory, computation, and evolving physical state become intertwined.
Work with mathematical structure.
Connect representations so mathematical objects can be constructed, explored, and inspected.
How we work
Think deeply. Build carefully. Make it understandable.
Begin with the question.
Make the problem precise enough to investigate. Connect the mathematical idea to the behaviour or experience it is meant to explain.
Learn through building.
Turn ideas into software and experiments. Inspect what happens, measure the relevant tradeoffs, and let the results shape the next iteration.
Make the work legible.
Use clear language, visual explanations, and inspectable systems so other people can understand the work and contribute to it.
Start a conversation
Good questions are a place to begin.
Bring a research question, a difficult workflow, or a mathematical idea that needs a better way to be explored. Tell us what you are working on and what you want to understand.
Explore a research question.
For researchers and builders interested in a shared problem, a useful experiment, or a new representation.
Discuss a collaborationEvaluate Dreamer in your work.
For teams exploring which context should reach an agent, how to measure the difference, and where Dreamer might fit.
Discuss your workflow