Synthetic Data Vault

Synthetic Data Vault is a Python library for teams that need synthetic versions of their tabular datasets. It learns patterns from real records and generates data that emulates them. Training and generation run on your own hardware with standard CPUs.

SDV handles single tables, connected relational tables and sequential data. Its model choices include GaussianCopula for statistical modeling and CTGAN for deep learning, so users can compare different approaches within the same library. Once trained, a synthesizer can generate additional data on demand.

The library also helps assess how closely the generated data resembles the original. Quality reports provide an overall score and detailed breakdowns, while visual comparisons help reveal differences between the real and synthetic datasets. These checks give teams a way to diagnose problems before relying on the generated records.

Users can adjust preprocessing, choose anonymization methods and define logical constraints that express business rules. Those controls matter when a dataset needs to preserve statistical patterns while respecting rules about which values can occur together. Anonymization is configurable; it isn't a blanket privacy guarantee.

SDV is source-available under Business Source License 1.1. Its additional use grant excludes operating a commercial service that exposes its synthetic-data functionality to third parties.