JFrog ML is JFrog Ltd's MLOps platform, born the Qwak acquisition, covering the machine learning and GenAI model lifecycle: training, versioning, deployment, serving and monitoring. It uses JFrog Artifactory as an immutable model registry and JFrog Xray to scan models and AI artifacts before deployment. It includes a Feature Store, an Inference Lake and serving via API endpoint, batch and streaming. Licensing combines ML Credits with a developer base.
JFrog ML is JFrog Ltd's MLOps platform for building, training, deploying, serving and monitoring machine learning models and GenAI applications, born the acquisition of Qwak AI — announced in June 2024 and completed the following month — and integrated into the JFrog Software Supply Chain Platform.
JFrog's thesis is to treat a model as a software package: instead of maintaining an MLOps pipeline separate the application pipeline, JFrog Artifactory acts as the immutable model registry, providing versioning, traceability and governance, while JFrog Xray scans the models, containers and artifacts for vulnerabilities and license issues before deployment. The practical result is that models and traditional artifacts live in the same registry, under the same access policy.
Architecturally, the platform separates the control plane the data and execution components: model registry and artifact storage, a Feature Store for feature data in training and real-time inference, an Inference Lake for prediction logs and operational metrics, and model serving on scalable endpoints.
Available scope depends on the contracted bundle:
Unified governance because the model enters the same registry as every other artifact — permissions, auditing and version history match the rest of engineering, with no second tool to control.
Model risk caught before it ships because JFrog Xray scans models and AI artifacts, including those sourced Hugging Face, on the same pipeline that examines traditional packages.
Viable rollback because each model version is an identifiable immutable artifact: reverting to the previous version is a registry operation, not an environment rebuild.
Less dependence on infrastructure engineering because serving, scaling and monitoring come ready — the data science team publishes the model without assembling a deployment pipeline scratch.
Consistency between training and inference because the Feature Store serves the same feature definition in both phases, eliminating the class of error where a model learns with one calculation and infers with another.
JFrog ML is sized for data science and machine learning engineering teams that already run models in production and face scaling problems, and for platform and security areas that need to bring AI assets into the existing SDLC.
Recurring scenarios: the company already on Artifactory that wants to avoid contracting a disconnected MLOps platform; the organization that must audit which models are in production and with which dependencies; and the team consuming open models public repositories that needs to scan them before use.
The core differentiator is convergence with artifact management: because the model registry is Artifactory itself, models inherit the versioning, access control and security scanning the organization already applies to software — instead of sitting in a parallel silo with its own governance.
The second differentiator is full lifecycle coverage under a single contract: training on CPU or GPU, Feature Store, serving, monitoring and retraining automation. The third is proximity to supply chain security — the same JFrog Xray that blocks a malicious npm package examines the model before deployment.
JFrog ML is not sold standalone: it is contracted as a bundle on top of a JFrog Artifactory subscription in the Enterprise X or Enterprise + editions. Unified MLOps is available on both; Ultimate MLOps is exclusive to Enterprise +.
In the managed model, JFrog indicates native support for AWS and Google Cloud for JFrog ML workloads — unlike the rest of the platform, which also runs on Microsoft Azure. Execution can happen on JFrog's infrastructure or the customer's own. Training and fine-tuning run on CPU or GPU machines depending on the workload.
The platform base is described on the JFrog Artifactory Cloud and JFrog Artifactory Self-Hosted pages, and model scanning on the JFrog Xray page.
JFrog ML is licensed on a dual metric: ML Credits, which measure platform processing consumption, plus the developer base of the contracted bundle — 200 on Unified MLOps and 500 on Ultimate MLOps. The model is an annual subscription, following the Artifactory contract.
Enterprise X and Enterprise + subscriptions already include a base amount of ML Credits. Consumption beyond that base does not block the account: it is logged and billed automatically based on usage, the same way platform storage and transfer consumption works. Because credits track processing, what drives the figure up is build time, training time and allocated machine resources — a model trained on GPU consumes differently a light CPU job.
There is no single list value because sizing changes the outcome: chosen bundle, developer base, estimated training and inference volume, the Artifactory edition in use, contract term and whether it is a new purchase or a renewal. To buy JFrog ML in Brazil, OSB Software maps that scope with you, sends the commercial proposal, issues the Brazilian invoice and delivers the licenses — request a quote to get the exact figure for your scenario.
Is JFrog ML the former Qwak? Yes. JFrog announced the Qwak AI acquisition in June 2024 and completed it in July 2024, and the platform is now sold as JFrog ML, integrated with Artifactory and Xray.
Do I need Artifactory to use JFrog ML? Yes. JFrog ML is contracted as a bundle on an Enterprise X or Enterprise + subscription, and uses Artifactory itself as the model registry.
Which clouds does JFrog ML run on? JFrog indicates native support for AWS and Google Cloud for JFrog ML workloads, running on JFrog's infrastructure or the customer's.
Does JFrog ML handle LLMs and GenAI, or only classic models? Both. The platform covers everything linear regression to deep learning and LLMs, including fine-tuning and embedding deployment.
Which bundle includes the Feature Store? Ultimate MLOps, alongside Multi-Environment Setup, advanced deployment strategies and Multi-Region Deployment.
OSB Software is an official JFrog Ltd partner in Brazil and supplies 100% genuine JFrog ML licenses to companies, with Brazilian invoicing, local contracting in Portuguese and full legal compliance — the safe way to buy imported software under a Brazilian corporate entity, with no risk of irregular licensing.
When you buy JFrog ML OSB Software, you get:
If you are looking for where to buy JFrog ML in Brazil with legal certainty and properly licensed imported software, request a quote: OSB Software delivers reliable technology, qualified support and a simple, transparent purchase.
You don't know what software you need or you haven't found what you were looking for? We have a team ready to help you choose the right software for your company.