ZenML

What Is ZenML?

The assumption in MLOps procurement is usually that bigger means safer: pick the platform with the most funding, the largest team, the longest feature list. ZenML challenges that logic on paper — it's a small Munich company, founded in 2021 by Hamza Tahir and Adam Probst, with roughly 21 staff and $6.4 million raised in total. Yet it sits on genuinely wide adoption because it solved a narrower, sharper problem than most well-funded rivals attempt: making ML pipelines portable across whatever orchestration and tracking tools a team already has.

Where platforms like Hopsworks or Valohai sell you their infrastructure, ZenML sells you a framework that sits on top of infrastructure you already own or plan to adopt. You write a pipeline once in Python. A pluggable "stack" system then lets you point that pipeline at Airflow, Kubeflow, SageMaker, or Vertex AI as the orchestrator. Swap in MLflow or Weights & Biases as the experiment tracker without touching the pipeline code itself. That's a genuinely different bet than most of the category makes, and it's the reason a team this size punches above its funding weight.

The company is registered as ZenML GmbH in Munich (Handelsregister HRB 268487), raised a €3.5 million seed round that later extended to €6 million ($6.4 million) led by Point Nine Capital with existing backer Crane Venture Partners. The open-source core is free and unlimited. A paid ZenML Pro tier adds a hosted control plane, model registry, and governance features for teams that want the framework's flexibility without running their own control plane.

Key Features

The stack architecture

ZenML's central idea is the "stack" — a configuration object where orchestrators, artifact stores, and experiment trackers are interchangeable components rather than baked-in choices. Run a pipeline locally during development, then point the same code at a Kubeflow cluster or SageMaker in production by swapping the stack configuration, not the pipeline logic. This matters for organisations mid-migration between cloud providers, or ones that refuse to commit their entire ML workflow to a single vendor's orchestration layer.

Pipeline orchestration and lineage

Every pipeline run automatically tracks its artifacts, metadata, and metrics, with full lineage from raw input data to a trained model. The Model Control Plane surfaces that lineage so a team can trace exactly which dataset, code version, and hyperparameters produced any given model in production. That's a genuinely useful capability when a model starts behaving oddly and someone needs to reconstruct how it got there.

Broad integration catalogue

ZenML integrates with more than 50 MLOps tools, spanning orchestrators, artifact stores, and experiment trackers, including a direct integration with MLflow rather than treating it as a competitor. For teams with existing MLflow experiment-tracking investment, ZenML wraps around it instead of asking you to rip it out — a meaningfully different pitch than platforms that expect a full-stack replacement.

Codespaces and Pro tier extras

The paid ZenML Pro tier adds Codespaces — hosted, cloud-based development environments — alongside the Model and Artifact Control Planes and priority support. These features target teams that have outgrown the self-hosted open-source setup and want a managed control plane without giving up the underlying framework flexibility that drew them to ZenML in the first place.

CI/CD integration

Pipelines integrate into standard CI/CD workflows, so a model retraining or validation pipeline can trigger from a Git push the same way a software deployment pipeline does. This closes a gap that exists in some MLOps platforms where experiment tracking and production deployment live in separate, poorly connected systems.

Pricing

The open-source tier is free, self-hosted, and genuinely unlimited — no execution caps, no project caps, which is unusual in a category where "free tier" often means "trial with a ceiling." That generosity is a deliberate open-source strategy: get teams comfortable with the framework before they ever consider paying for it.

The jump to paid tiers is steep, though. ZenML Pro's Scale tier starts at $999 per month for just 2,000 pipeline executions and 3 projects — a significant leap from free with no mid-priced tier bridging the gap. Teams outgrowing the free tier's practical limits but not yet ready for a thousand-dollar monthly commitment sit in an awkward middle. Enterprise pricing is custom and adds SSO, audit logs, and air-gapped deployment for regulated environments, but again requires a sales conversation rather than self-serve signup.

EU Compliance and Privacy

ZenML GmbH is headquartered in Munich, Germany, and registered under German company law (HRB 268487), placing it fully under GDPR and EU jurisdiction. Because the open-source framework is entirely self-hostable, organisations with strict data residency requirements can run pipelines, artifact stores, and metadata tracking completely within their own EU infrastructure, with zero dependency on ZenML's hosted services.

The paid Pro tier's hosted control plane introduces a third-party dependency that self-hosted deployments avoid. Regulated teams should confirm hosting location and data-processing terms before adopting Pro rather than assuming EU hosting by default. An Enterprise-tier air-gapped deployment option addresses the most stringent cases directly.

Who It's Best For

Platform engineering teams building on existing infrastructure who need a pipeline layer that doesn't demand ripping out an orchestrator or tracker they've already invested in. If you already run Airflow or use MLflow for tracking, ZenML slots in rather than competing with them.

Cost-conscious teams starting out should lean on the free, unlimited open-source tier for as long as it holds — it's more generous than most category peers.

Regulated organisations needing full data control benefit from the framework's complete self-hostability, keeping every pipeline run inside their own infrastructure.

Teams wanting a fully managed platform rather than a framework to configure themselves may prefer Valohai or Hopsworks. Both ship proprietary, managed infrastructure instead of asking you to assemble your own stack.

The Verdict

ZenML makes a contrarian bet in a category where scale usually wins the marketing argument: a small, independent Munich team building a framework rather than a platform, and betting that flexibility beats breadth. It works, within limits. The open-source core is genuinely useful and unencumbered, the stack architecture solves a real portability problem, and the Munich HQ gives EU buyers a straightforward compliance story. Set against that is a company running on $6.4 million and 21 people, a Pro pricing ladder with an awkward gap between free and $999 a month, and a feature set that assumes you're comfortable configuring infrastructure rather than renting a finished platform. Teams that value control over convenience will find that trade works. Those who want MLOps as a managed service rather than a framework to assemble should look instead toward the broader AI developer tools category or the MLflow alternatives page for platform-style options.

Frequently Asked Questions

Is ZenML free to use?

Yes. The open-source core is free and self-hostable with unlimited pipeline executions and projects. A paid ZenML Pro tier adds a hosted control plane starting at the Scale tier ($999/month for 2,000 executions), with custom Enterprise pricing for larger teams.

Is ZenML GDPR compliant?

Munich-registered ZenML GmbH (Handelsregister HRB 268487) is subject to German and EU data protection law. Because the framework is fully self-hostable, teams needing complete data residency control can run it entirely on their own EU infrastructure.

How does ZenML compare to MLflow?

MLflow is primarily an experiment-tracking and model-registry tool. ZenML is a pipeline-orchestration framework that can use MLflow as one component in its stack, so the two are often complementary rather than direct substitutes.

How does ZenML differ from Valohai or Hopsworks?

Valohai and Hopsworks are managed platforms with proprietary infrastructure and no comparable open-source core, unlike ZenML's framework-first, open-source Python library that you install into your own code. Teams that want to own their stack lean toward ZenML, while teams that want a managed service lean toward Valohai or Hopsworks.

How big is the company behind ZenML?

ZenML GmbH is a small company, founded in Munich in 2021, with roughly 21 staff as of May 2026 and $6.4 million raised in total across a seed round led by Point Nine Capital with Crane Venture Partners.