TokenTown: A visual way to understand how LLMs work

TokenTown is an interactive visual simulation that demonstrates how large language models work by representing each stage of a transformer architecture as districts in an isometric city. The tool runs genuine computations in the browser—including tokenization, embedding, attention mechanisms, and sampling—with a miniature model to show how tokens flow through layers and generate text one word at a time. While the underlying weights are random (producing nonsensical output), the visualization accurately illustrates the real mechanical processes that enable LLMs to function.