Hopsworks Feature Store

Overview

Hopsworks Feature Store is a platform for managing, reusing and serving features in machine-learning workflows. It is best suited to teams building production AI that need online and offline storage, feature governance and a choice of managed or self-hosted deployment. Its breadth is a strong fit for shared, operational feature work, but the free plan’s one-project limit makes it restrictive for teams managing multiple projects.

Key features

Hopsworks combines an offline store using Apache Hudi tables on HopsFS with an online RonDB key-value store. That pairing supports both historical training workflows and online feature access, which is useful for teams that need to develop models and serve features from one platform. Point-in-time training datasets reflect feature values at a historical moment, helping prevent data leakage during training.

The Feature Registry provides metadata search, automatic feature analysis and statistics, version comparison and access control. These tools help teams find and govern shared features as definitions evolve. Monitoring tracks feature distributions, detects drift and can send alerts, giving production teams a way to identify changes in the data their models depend on.

Integrations include Airflow, Apache Flink, Databricks, Snowflake, Spark, AWS SageMaker, Vertex AI and Weights & Biases. The range can suit teams already working across data pipelines and model platforms, though the value depends on their stack matching those integrations.

Security capabilities include encryption at rest and in transit, project-based access controls, and authentication options such as two-factor authentication and SSO. Hopsworks states that it is certified under ISO 27001 and SOC 2 Type 2 and is GDPR compliant. These are relevant credentials for organizations with security and compliance requirements, alongside Enterprise’s dedicated support and guaranteed SLA.

Pricing

Hopsworks offers a Free plan at no cost. It includes one project, the Feature Store and Model Registry, and community support; no credit card is required. That is a useful starting point for an individual or small team assessing the core workflow, but one project is a hard ceiling for broader use.

The SaaS plan is billed Pay as you go, with custom pricing. It includes unlimited projects, the Feature Store and Model Registry, Model Serving, usage-based pricing, community support and a platform SLA. It removes the Free plan’s project cap and adds serving, making it the more suitable option for teams that need a managed service across multiple projects. Enterprise has custom pricing and includes all SaaS features, on-premise and air-gapped deployments, a dedicated support team, custom integrations and a guaranteed SLA. It fits organizations with deployment, support or service commitments beyond the SaaS offer.

Platforms

Hopsworks supports API, Linux, self-hosted and web use, and offers both online and offline stores, point-in-time joins, and managed and self-hosted deployment. Deployment choices include managed SaaS, Kubernetes on AWS, Azure or GCP, and air-gapped data centres. Self-hosting is a substantial infrastructure commitment: the installer requirements name AWS EKS, Google GKE, Azure AKS and OVHCloud, require Kubernetes 1.27 or higher, and recommend at least 4–5 nodes. That flexibility is valuable when control over deployment matters, but less appealing to teams without Kubernetes capacity.

Who it's for

Hopsworks presents its platform for data scientists, data engineers, developers, platform engineers, security engineers and administrators building production AI systems. It is a strong candidate for cross-functional teams that need feature reuse, historical training data, monitoring and managed or controlled deployment options. It is less compelling for a small, single-project effort that does not need this breadth, or for teams unable to operate the infrastructure required for self-hosting.

Pros and cons

• Pro: Online and offline stores plus point-in-time training datasets cover both serving and historical model-building needs.
• Pro: Registry and monitoring tools support feature discovery, version comparison, access control and drift alerts in production workflows.
• Pro: SaaS, cloud Kubernetes and air-gapped deployment options serve teams with different hosting and control requirements.
• Con: Free is limited to one project, so it does not accommodate a growing multi-project workload.
• Con: SaaS and Enterprise both have custom pricing, and self-hosting calls for Kubernetes 1.27 or higher with at least 4–5 nodes recommended.

Alternatives

Feature Store Software is the broader category to explore when comparing feature-store options.

Choose Canal if a free, open-source project with API, Linux, macOS and self-hosted support better fits your needs. Chronon is another free Apache 2.0 option, with API and self-hosted platforms and no usage limits stated. Feast offers a free open-source feature store across API, Linux, self-hosted and web platforms, while OpenMLDB is a free open-source machine-learning database with standalone and cluster versions. Feathr is also free and supports API, Linux, self-hosted and web.

Chalk is a paid alternative with API, self-hosted and web platforms. Databricks Feature Store may suit teams already considering Databricks: its pay-as-you-go plan bills monthly by usage with per-second granularity, no upfront costs, and rates that vary by product, cloud provider and region.

Verdict

Hopsworks is a strong choice for production AI teams that need a governed feature workflow spanning point-in-time training, online serving, monitoring and flexible deployment. Its main advantage is that breadth across storage, registry and operations; its main drawbacks are custom pricing for paid plans, the Free plan’s single-project cap and the infrastructure needed to self-host. Choose it when those capabilities justify the commitment, and look elsewhere for a simpler free option or a service whose pricing terms are more explicit.