TensorSpace is an open-source JavaScript framework for building interactive 3D visualizations of neural networks that run entirely in the browser. Developed by the TensorSpace team and published on GitHub, where it has earned more than 5,200 stars, it lets you load pre-trained deep learning models and explore their inner structure layer by layer in an intuitive three-dimensional scene.
Most deep learning models are black boxes: tensors go in, predictions come out, and the transformations in between stay invisible. TensorSpace attacks that problem with visualization. Built on Three.js for rendering and TensorFlow.js for in-browser inference, it converts models from TensorFlow, Keras and TensorFlow.js into navigable 3D objects. You can rotate a network, zoom into individual layers, inspect feature maps, and watch how an input image is progressively transformed as it flows through convolutions and pooling stages. The project was last updated in 2022 and remains a widely referenced teaching and demonstration tool.
Because the visualization runs entirely client-side, TensorSpace pages can be embedded in documentation sites, course material and blog posts, letting readers manipulate a real, working model instead of looking at static diagrams of one.
TensorSpace is completely free and open source under the Apache-2.0 license. There is no hosted version, no premium tier and no usage limit: you download the library, plug in your model and host the visualization wherever you like.
TensorSpace suits machine learning educators, students and researchers who need to explain or present how a neural network works, as well as developers building interactive demos that make model behavior tangible for non-technical audiences.