Nooks makes interactive worlds for people and AI. Build brick creations you can hold, make games with intelligent characters, and explore environments that help machines learn to act.
We started Nooks with a simple idea: the best way to understand something is to build it. Today, we are making tools for people and AI to create, learn and solve problems together.
Machines learned a lot from what we wrote down. But much of what makes people capable comes out in action: building, playing, finding the way somewhere new, making a mistake and fixing it.
Worlds make that action visible. A game or a construction has a state, rules, choices and consequences, so it shows not just what someone knows, but whether they can act on it. Nooks makes these worlds easier to build, host, play in and study.
People should want to build in Nooks even if they never think about AI. What they make and do, shared only with their express permission, helps AI get better at acting in worlds. Better AI then makes the tools better for people.
The human side of Nooks: make objects, places and experiences, play them, and share them. Everything is built on one shared representation, so a person and an AI agent can work on the same world.
A picture in, a real brick-built set out, with a 3D preview, parts list and digital manual.
Live nowOpen any set, edit it brick by brick, play its steps and export it. Where people and AI agents edit the same construction.
Live nowReal cities and buildings rebuilt from open map data, floor plans and photos, ready to explore, edit and build in.
In developmentAn API for putting agents inside games and worlds: they see, remember, pursue goals and act through the moves the world allows.
In developmentGames and experiences with characters that remember you and respond to the world around them.
In developmentPublish, remix and sell designs, kits and worlds, with credit for everyone who built on them.
PlannedGames and constructions are small worlds with rules and consequences. We turn them into environments and data that make AI agents measurably better at doing things, not just talking about them.
Winning a game or finishing a build takes planning, memory, spatial reasoning and working with others.
A part connects or it doesn’t. A game is won or lost. Checkable outcomes are what AI improves on fastest.
Whether a skill learned in one world helps in another is a question we test, not something we assume.
Agents act inside worlds with real rules: placing parts in a construction that has to hold, or playing games that reward long-horizon plans. Every attempt is scored, and we say which scores are checked and which are judged.
How people really build and play: designers fixing a model, players finding their way through a game. Recorded only with consent and, for games, the publisher’s permission.
We’re working toward a way for people who create or meaningfully remix a kit and share it publicly, such as on X, to take part in a data-sharing programme. With their permission, the steps they take to build could help train AI. If that data is ever licensed commercially, we hope to share some of the revenue with eligible contributors. Sharing would be required to qualify; the quality of a contribution would help determine any compensation. The programme is an idea we’re working toward, not a payment available today.
Environments to train and test agents, evaluations you can reproduce, and permitted human demonstration data.
Talk to usTurning a picture of your dog into a kit you can build is a hard reasoning problem: read the picture, keep what makes it recognisable, translate curves into a finite set of parts, balance likeness against strength and part count, plan the order to build it, and end with something real. That makes it a great first product and a great first test.
Every world we make has to be worth using for its own sake. Research value comes after.
A build is done when software has checked it holds together with parts that exist. We say which signals are checked and which are judged.
Every model records what came from a picture and what we had to imagine. We never pass a guess off as fact.
Research uses only what people and partners have agreed to share, kept apart from ordinary design projects.
Our first product, live today: something you love, in real bricks.
Corgi · open the demo
Tell us what you are reaching out about.
Or email us directly
Developers can start from the developer tools.
