secretflow

Why we included this project

Training models or running analytics on data that lives with different institutions usually means either pooling raw records or shipping them around. SecretFlow takes the other route: it keeps each party's data in place and runs the computation over cryptographic protocols that protect the underlying values. The framework hides secure multi-party computation, homomorphic encryption, differential privacy, and trusted execution environments behind a device abstraction layer, so you write the computation as a dataflow and let SecretFlow decide which protection applies. It supports both horizontally and vertically partitioned data, which covers federated learning and split learning, and the workflow runs from data processing through model training to hyperparameter tuning. Companion libraries like Kuscia for orchestration and SPU and HEU for secure computation round out the stack, which matters when you are planning a real deployment rather than a quick experiment.