Skip to the content
Know your tech before you buy it. Makers: get listed from $49/mo

OpenMLDB

openmldb.ai

  • Free plan
  • macOS
  • Linux
  • API
  • Self-hosted
7.0 Club score Visit openmldb.ai

What OpenMLDB is

OpenMLDB is an open-source machine learning database and feature platform designed to keep features consistent between model training and inference. Teams use SQL to write feature-engineering scripts, deploy them online, and configure online data sources. Its architecture combines real-time and batch SQL engines with a unified execution-plan generator; the documentation says the real-time engine can produce features in a few milliseconds. SQL extensions include LAST JOIN and WINDOW UNION. OpenMLDB offers a cluster edition for large production applications and a single-node standalone edition for evaluation or demonstration. Listed production capabilities include distributed storage and computing, fault recovery, high availability, scale-out, upgrades, monitoring, and heterogeneous memory support. Integrations can bring Apache Pulsar streams into OpenMLDB and connect feature engineering to DolphinScheduler workflows. Kubernetes deployment covers its offline and online engines, but the documented cluster setup lacks a TaskManager, so LOAD DATA, SELECT INTO, and offline-related functions are unsupported there. OpenMLDB is free and self-hosted, with online and offline stores, point-in-time joins, feature monitoring, and both serving modes.

Who it suits

OpenMLDB suits teams building SQL-based feature engineering workflows for machine learning, from standalone evaluation to larger self-hosted production deployments. Its tools may also fit teams using Pulsar, DolphinScheduler, or Kubernetes.

Good to know, and what to watch

  • Free and open source.
  • Supports online and offline stores.
  • Provides point-in-time joins and feature monitoring.
  • Offers standalone and cluster deployment options.
  • Includes batch and real-time SQL engines.
  • Kubernetes cluster deployment lacks a TaskManager.
  • Several offline functions are unsupported in that Kubernetes deployment.
  • Real-time feature timing is described as a few milliseconds, not a guarantee.

OpenMLDB

Free

Open source

Getting started

  1. Open the OpenMLDB website
  2. Choose the standalone version for evaluation or the cluster version for production
  3. Deploy the offline and online engines with Kubernetes, if applicable
  4. Use the Spark distribution through its Scala, Java, Python, or R interface
  5. Connect Apache Pulsar or DolphinScheduler when those workflows are needed

Limits to check

In the documented Kubernetes cluster deployment, no TaskManager is included. As a result, LOAD DATA, SELECT INTO, and offline-related functions are unsupported there.

The Club's review

OpenMLDB is an open-source machine-learning database and feature platform for teams that need consistent features across training and inference. It suits engineering groups comfortable operating self-hosted infrastructure and expressing feature logic in SQL. The combination of online serving, offline processing and production controls is substantial, but Kubernetes deployments have a material offline limitation.

Overview

OpenMLDB makes SQL the main path from feature engineering to serving: teams write feature scripts, deploy them online and configure online data sources in the same platform. A real-time SQL engine is paired with a batch SQL engine built on a tailored Spark distribution, while a unified execution-plan generator keeps the two execution paths aligned. That design is useful when training and inference must use consistent definitions rather than separate implementations.

The real-time engine is documented as producing features in a few milliseconds, making it a plausible choice for latency-sensitive inference. This is a documented capability, not a promise for every query or deployment. A lightweight standalone build is intended for evaluation and demonstrations; the cluster edition is aimed at large-scale production and requires the operational ownership that comes with self-hosting.

Key features

  • SQL feature engineering: Extensions including LAST JOIN and WINDOW UNION provide constructs specifically for feature work. SQL-oriented teams can keep feature definitions readable and versionable, while newcomers will need to learn OpenMLDB's additional syntax.
  • Online and offline stores: The platform provides both stores, point-in-time joins and feature monitoring. That covers the core needs of training-data correctness and serving, instead of forcing separate systems for historical and online features.
  • Unified execution: Real-time and batch SQL engines share an execution-plan generator. The tailored Spark distribution also exposes Scala, Java, Python and R interfaces; its precompiled AllinOne build supports Linux and macOS.
  • Production operation: Distributed storage and computing, fault recovery, high availability, scale-out, upgrades, monitoring and heterogeneous memory support are all part of the documented architecture. These capabilities make the cluster version more than a development store, but they also imply a meaningful administration burden for smaller teams.
  • Integrations: The Pulsar Connector imports real-time streams from Apache Pulsar. A DolphinScheduler task supports offline import, feature extraction, SQL deployment and online import, which helps place feature work inside scheduled workflows.
  • Kubernetes deployment: Deployment guidance covers online and offline engines and is tested with Kubernetes 1.19 or later and Helm 3.2.0 or later. However, the documented cluster deployment has no TaskManager, so LOAD DATA, SELECT INTO and other offline-related functions are unsupported there. Teams depending on those operations should use another deployment approach or accept a significant gap.

Pricing

OpenMLDB is free and open source. Its sole listed plan, OpenMLDB, costs free and includes standalone and cluster versions. There are no seat charges, usage quotas or trial expiry to plan around in the published terms. The trade-off is that self-hosted users supply the infrastructure, deployment work and ongoing operations themselves; there is no managed-service tier in this plan.

Platforms

OpenMLDB supports API access, Linux, macOS and self-hosted deployment. Serving modes include both online and batch operation. The standalone build is suited to local evaluation, while production-scale use is centered on the cluster version. Kubernetes is supported with the stated version requirements, subject to the missing TaskManager and its offline-function restrictions.

Who it's for

Choose OpenMLDB if your machine-learning team wants one SQL-centered workflow for feature definitions, historical training data and low-latency serving, and has the skills to run distributed infrastructure. It is particularly relevant to organizations already using Spark, Pulsar or DolphinScheduler, or those needing point-in-time joins and monitored features under their own control.

It is a weaker fit for teams that want a fully managed feature service, minimal operations, or Kubernetes-only workflows that rely on LOAD DATA, SELECT INTO or other offline functions. The standalone version gives evaluators a sensible starting point, but production adoption still means running and maintaining the platform.

Pros and cons

Pros

  • One SQL workflow across training and inference: Shared planning and paired engines reduce the conceptual split between batch feature creation and online serving.
  • Strong self-hosted production foundation: High availability, fault recovery, scale-out, monitoring and distributed storage address concerns that basic feature-store projects may leave to the operator.
  • Useful ecosystem hooks: Pulsar and DolphinScheduler integrations connect streaming and scheduled feature pipelines to the platform.
  • No license cost: The open-source plan includes both standalone and cluster editions at no cost.

Cons

  • Kubernetes offline gap: Without a TaskManager, documented Kubernetes cluster deployments cannot run LOAD DATA, SELECT INTO or other offline-related functions.
  • Self-hosting responsibility: Infrastructure, upgrades, monitoring and reliability remain the user's job rather than a provider's.
  • Specialized SQL learning curve: LAST JOIN and WINDOW UNION add power for feature engineering but require familiarity beyond standard SQL.

Alternatives

Feature Store Software is the broader category overview for readers comparing multiple approaches.

  • Amazon SageMaker Autopilot is the better direction for a paid, API-and-web service with a free trial and pay-as-you-go SageMaker pricing.
  • Canal suits readers wanting a free Apache-2.0 project with the same API, Linux, macOS and self-hosted reach.
  • Feast is another free, open-source feature store available through API, Linux, self-hosted and web environments.
  • Hopsworks Feature Store offers a free one-project tier with Feature Store, Model Registry and community support, plus a custom-priced enterprise plan.
  • JFrog ML Feature Store is a paid, self-hosted or web option with a free trial and sales-led pricing.
  • Databricks Feature Store fits teams preferring a managed web/API service billed monthly by usage, with rates varying by product, cloud and region.
  • Red Hat OpenShift AI Feature Store is for organizations standardizing on OpenShift and subscription-based core-pair or bare-metal support.
  • Chronon is a free Apache 2.0 open-source alternative with API and self-hosted deployment.

Verdict

OpenMLDB is a strong choice for self-hosting teams that want SQL-based feature engineering with both online and offline stores, point-in-time joins and production-grade distributed capabilities at no license cost. Look elsewhere if you need managed operations or depend on offline functions in the documented Kubernetes deployment, where the missing TaskManager is a decisive constraint.

The verdict

OpenMLDB brings SQL-based feature engineering and online and batch processing into a self-hosted platform. Check the Kubernetes TaskManager limitation if your deployment depends on offline-related functions.

Questions people ask

Is OpenMLDB free?

Yes. It is open source, has a free plan, and the OpenMLDB plan is listed as free.

What deployment versions are available?

OpenMLDB provides a cluster version for large-scale production applications and a lightweight standalone version for evaluation and demonstration.

Does it support online and offline serving?

Yes. It provides online and offline stores, and its serving modes are both online and offline.

Can OpenMLDB run on Kubernetes?

Yes. Deployment documentation covers both its offline and online engines. The documented tool is tested with Kubernetes 1.19 or later and Helm 3.2.0 or later.

What integrations are available?

The project provides a connector for Apache Pulsar real-time streams and a DolphinScheduler task for feature-engineering workflows.

What the maker publishes

Batch and real-time engines
Its architecture includes a real-time SQL engine, a batch SQL engine based on a tailored Spark distribution, and a unified execution plan generator. source
Community support
The project directs users to GitHub Issues for bug reports and feature requests, GitHub Discussions, Slack, and a developer mailing list. source
Deployment options
OpenMLDB has a cluster version for large-scale production applications and a lightweight single-node standalone version for evaluation and demonstration. source
DolphinScheduler integration
OpenMLDB provides a DolphinScheduler task for integrating feature engineering into workflows, including offline import, feature extraction, SQL deployment, and online import. source
Kubernetes deployment
The deployment guide describes Kubernetes deployment for both OpenMLDB's offline and online engines. source
Kubernetes limitation
The documented Kubernetes cluster deployment does not include a TaskManager, so LOAD DATA, SELECT INTO, and offline-related functions are unsupported in that deployment. source
Kubernetes requirements
The Kubernetes deployment tool is tested with Kubernetes 1.19 or later and Helm 3.2.0 or later. source
Production capabilities
The documentation lists distributed storage and computing, fault recovery, high availability, scale-out, upgrades, monitoring, and heterogeneous memory support. source
Pulsar integration
The OpenMLDB Pulsar Connector is described as a way to import real-time data streams from Apache Pulsar into OpenMLDB. source
Real-time features
The documentation says its real-time SQL engine can produce features in a few milliseconds. source
Spark distribution
The OpenMLDB Spark distribution provides Scala, Java, Python, and R interfaces, and its precompiled AllinOne version supports Linux and macOS. source
SQL extensions
OpenMLDB extends SQL for feature engineering with syntax including LAST JOIN and WINDOW UNION. source
SQL workflow
OpenMLDB uses SQL to develop feature engineering scripts, deploy them online, and configure online data sources. source
What it does
OpenMLDB is an open-source machine learning database and feature platform for consistent features in training and inference. source

Checked against the maker's pages on 3 Oct 2026.