Easy enough for developers new to AI yet powerful and customizable enough for data scientists to replace their entire predictive-modeling workflow. Kumo connects directly to your data source for reads and writes — no pipelines required.
Connect your data sources or upload files, choose tables, and confirm links between tables. Kumo works directly with relational data where it lives — no need to build pipelines, have graph data, or know anything about graph transformers.
Docs on data connectorsKumo models use deep learning to determine what attributes and relationships matter, so there's no feature engineering or extensive data prep involved.
Learn how: Relational Deep Learning (RDL)Choose the tables to train on, describe the predictive task, and your model will be ready between a few minutes to a few hours. Under the hood, Kumo creates a graph from your relational data and uses graph-based deep learning to train both predictive and embedding models.
Model training docsValidate the model on unseen holdout data to ensure real-world performance. Kumo automatically splits your historical data—reserving the latest window for unbiased testing — so you get clear, task-specific metrics such as accuracy, AUROC, and RMSE. Uncover insights, fine-tune your approach, and deploy with confidence.
Evaluation docsSee exactly what your model is learning. Quickly identify key contributors in your data. Confidently explain model behavior to your team. See what needs changing and iterate fast.
Explainability docsStart or schedule model runs to get prediction values or embeddings based on the latest data. Results go straight to your data warehouse, database, or S3 bucket for instant access by your apps and processes.
Model runs docsConnect your data sources or upload files, choose tables, and confirm links between tables. Kumo works directly with relational data where it lives — no need to build pipelines, have graph data, or know anything about graph transformers.
Kumo models use deep learning to determine what attributes and relationships matter, so there's no feature engineering or extensive data prep involved.
Choose the tables to train on, describe the predictive task, and your model will be ready between a few minutes to a few hours. Under the hood, Kumo creates a graph from your relational data and uses graph-based deep learning to train both predictive and embedding models.
Validate the model on unseen holdout data to ensure real-world performance. Kumo automatically splits your historical data—reserving the latest window for unbiased testing — so you get clear, task-specific metrics such as accuracy, AUROC, and RMSE. Uncover insights, fine-tune your approach, and deploy with confidence.
See exactly what your model is learning. Quickly identify key contributors in your data. Confidently explain model behavior to your team. See what needs changing and iterate fast.
Start or schedule model runs to get prediction values or embeddings based on the latest data. Results go straight to your data warehouse, database, or S3 bucket for instant access by your apps and processes.
Kumo is priced as a flat fee based on expected usage, deployment model, and support level. Contact us to discuss your use case and get an accurate estimate.
Talk to usKumo is available as a fully managed cloud service, a Snowflake Native App, and a Databricks Lakehouse App. Learn more about deployment options.
Kumo is SOC 2 Type 2 compliant, adheres to HIPAA and GDPR directives and built using industry standards like NIST 800-53, NIST 800-53, and ISO 27000 series. Learn more about security at Kumo.
