Zetane Viewer is a free, open-source desktop tool on GitHub for visualizing machine learning models and their internal tensors in 3D. With about 1,900 stars, it acts like a neuroimaging tool for artificial neural networks, opening up the black box so engineers can literally see how a model is built and what happens inside it.
The viewer loads model files, including ONNX models and Zetane’s own ZTN format, and renders the network architecture as an explorable 3D scene. Users can navigate layers, inspect weights, biases, feature maps and layer output tensors, and compare activations, which makes debugging architectures and explaining model behavior far more intuitive than reading raw numbers. Installers are provided for Windows, Linux and Mac, and a Python API lets you launch a Zetane workspace directly from existing scripts or notebooks. Although the repository’s last release dates to 2022, it remains a distinctive, approachable entry point into neural network interpretability tooling.
Zetane, the Montreal company behind the viewer, positioned it as the entry point to its broader neural network understanding platform, and the free viewer has been used in courses and meetups to give newcomers an intuitive picture of tensors flowing through a network.
Zetane Viewer is free. The installers linked from the GitHub repository can be downloaded at no cost, making professional-style model visualization accessible to students, researchers and engineers without a software budget.
Zetane Viewer is for ML engineers, researchers and educators who want to understand, debug and demonstrate neural networks visually, from checking tensor shapes in a new architecture to showing students what actually happens inside a trained model.