Manifold
Making agent work visible.
A visual workspace for coordinating agents, tools, and people.
Experimental
Explore the researchLoading ATMAI
Atmai is a research lab working across intelligence, computation, and mathematics. We turn questions into experiments, open tools, and products.

Questions worth working on
How intelligence remembers, learns, and becomes useful.
Dreamer
Dreamer constructs task-specific context for AI agents from information whose sources and revisions remain inspectable. It brings structure to changing decisions, competing accounts, and the information a task depends on.
Illustrative example
Follow a decision as it changes.
Launch the pilot on Friday.
Invite the existing pilot group.
Trace it to the source
Launch the pilot on Friday.
Invite the existing pilot group.
Launch date superseded
Move the pilot to Tuesday.
Keep the original scope.
Trace it to the source
Launch the pilot on Friday.
Invite the existing pilot group.
Launch date superseded
Move the pilot to Tuesday.
Keep the original scope.
Task brief
Prepare a launch update for the existing pilot group. Use Tuesday as the launch day.
Trace it to the source
Our programmes investigate how systems retain knowledge, coordinate work, learn from experience, and help people reason. Each begins with a question we can make concrete.
Making agent work visible.
A visual workspace for coordinating agents, tools, and people.
Experimental
Explore the researchExploring how systems learn.
Research into adaptive, continuously evolving computational substrates.
Research programme
Explore the researchWolfbone
Follow how mathematical ideas connect, inspect their sources, and build with them. Wolfbone brings an explorable mathematics library together with a visual workspace for finite collections and functions.
In development
Explore Wolfbone1 maps to a. 2 maps to b. 3 maps to a.
Explore our open-source experiments. Read the code, inspect the methods, and take the work further.
ATMAI on GitHubAn open research lab for studying, benchmarking, and extending adaptive graph learning.
We are a research lab that builds. Mathematics gives us a language for precise questions. Experiments challenge our ideas. Software lets people explore what those ideas can become.