Better model architectures.

Conway Labs researches the architectures, data, and engineering methods that can make capable AI smaller, more efficient, and practical to run.

Our story

Intelligence needs more than scale.

Our mission

Better model architectures.

A common reading of scaling laws is that more data and more compute will keep producing more intelligence. Scale has moved the field forward, but we do not believe it is the whole story.

TinyStories helped demonstrate that carefully designed, high-quality data can unlock surprisingly coherent behavior in very small language models. Across the field, smaller and more specialized models have also rivalled or outperformed much larger systems on focused tasks.

That points us toward a different operating belief: better architectures, good data, and disciplined engineering can produce more useful intelligence per parameter, per watt, and per device.

AI that fits where people use it.

Most leading models are still too large and resource-intensive to run locally on everyday consumer devices. That limits privacy, access, latency, and the kinds of products people can build around them.

We explore model architecture and systems engineering together, with the goal of making excellent AI efficient enough to run closer to the people and devices it serves.

Research tools

Conway Canvas is our first product.

Progress in architecture needs better ways to think, experiment, and communicate. Conway Canvas gives AI research scientists and engineers a visual workspace for architecting neural networks, understanding their structure, and turning designs into usable code.

Open Conway Canvas
The Conway Canvas visual model architecture canvas

Team

Building for the next architecture breakthrough.

A small team working across model research, developer tools, and efficient systems.

Stop typing your model. Start drawing it.

Conway Canvas is free. Open the canvas and build.