Maximize abstraction from infrastructure, provisioning and management. Hydrosphere Serving deploys models eg. on Local, AWS, Kubernetes, Mesos and YARN environments.
Everything is opensource. No premium features and enterprise versions. No hidden costs and learning curve can be as smooth as it gets.
No coding and no dependencies
Hydrosphere Serving needs just models metadata and runtime version to compose a deployable API instance.
ML Frameworks agnostic
Hydrosphere Serving cluster won’t lock you in to a specific machine learning framework, but lets you publish your data science toolbox by deploying scikit-learn, TensorFlow, mllib, deeplearning4j and other popular or custom machine learning runtimes in one click.
No training constraints
Flexible training of the data model within your favourite workspace (notebook) environment or through automatic training pipelines.
Native Spark ML support
Hydrosphere Serving natively supports Apache Spark prediction pipelines – no conversion to PMML is required.
Models versioning and A/B testing
Version control of predictive models included, to facilitate early and efficient comparative (A/B) testing.
Sidecar in action
Hydrosphere Serving has fully Sidecar compliant architecture to keep all runtime parameters (eg. networking) separated from the data science logic and component hierarchy.