PyG - Kumo



The open source framework founded by members of the Kumo team


Pytorch Geometric, the open source framework founded by members of the Kumo team, is the leading standard for building and training Graph Neural Networks.

PyG gives you full customization and flexibility to build the best models from the ground up. Built with the design principles of PyTorch, PyG makes it easy to build for any application, whether you are a machine learning engineer, a data scientist, or a researcher.

To learn more, see


Kumo builds on top of the PyG framework in a single highly reliant, scalable and performant solution out of the box. Connect your data sources and instantly run our sophisticated machine learning platform to query the future. With Kumo, you have full automation and the fastest time-to-value in making predictions.

What Should I Use?


I am a Data Scientist, Machine Learning Engineer, or a Researcher using data to solve problems

  • You can build state-of-the-art GNN models using the industry-leading framework for any use case
  • You can leverage optimized GNN-based operations, a toolbox of application-specific libraries and models, and frameworks for finding the best model architecture and parameters
  • You have highly customizable interfaces to enable any backend infrastructure


I am a Data/Business Analyst (marketing, growth, security), or a Data Scientist

  • You have business data and want to make predictions about the future based on that data, regardless of whether machine learning infrastructure is set up at your company
  • You can leverage a fully automated and highly optimized machine learning capabilities that are robust, performant, and scalable, out-of-the-box
  • You want fastest time-to-value for predictive analytics

PyG vs Kumo feature comparison

Feature PyG KUMO
Fully Automated state-of-the-art machine learning capabilities out of the box
Performance and auto-scaling to any size
Monitoring, Evaluation, and Explainability out of the box
Orchestration and cloud infrastructure
Automatically assembles the graph and pre-process the data
UI and API interfaces to make predictions using query language
Simple, unified APIs for building your own state-of-the-art graph networks
Flexible and customizable interfaces and libraries to build for any application
Model experimentation framework for finding the best model and parameters
Representation learning rather than manual feature engineering for building predictive models
ML that adapts to the shape of your data, without requiring pre-flattened tabular data

Query the future

Unleash the predictive power of your enterprise data

Request access