ZenML
Open-source AI platform that runs ML pipelines and agents on any stack, tracks runs and artifacts, and offers a managed Pro tier
Overview
ZenML describes its open-source project as "one AI platform from pipelines to agents." You write pipelines in Python and ZenML runs them on a "stack" of your choice (local, Kubernetes, cloud orchestrators), while recording every run, artifact and model version. In 2026 the company pushed hard into agent operations: its homepage leads with "AI that keeps getting better," it ships Kitaru for importing agent traces from Langfuse, LangSmith or Braintrust and replaying changes against past sessions, and ZenML Pro adds AI "engineers" for evaluation, reliability and cost work. ZenML GmbH is based in Munich, claims SOC 2 Type II and ISO 27001, and the latest open-source release is 0.98.1 (November 4, 2026).
The Verdict
Who Should Use ZenML?
Best For
- Teams that want portable Python pipelines they can move between clouds and orchestrators
- Groups running both ML pipelines and LLM agents who want one place for lineage
- Regulated companies that need on-prem or air-gapped deployment
Not Ideal For
- Teams that only need an experiment dashboard
- Small teams that need SSO or audit logs on a budget (Enterprise only)
What's Great
- Stack-agnostic: swap orchestrators and artifact stores without rewriting pipelines
- Wide integration list, including MLflow, Phoenix, LangGraph, PydanticAI and OpenAI Agents SDK
- Frequent releases and an Apache-2.0 core
- Named customers include AXA, JetBrains, Airbus Defence and Space, IKEA and Vodafone
Watch Out For
- Steep learning curve once you move past the tutorials
- Self-hosted upgrades take work
- SSO, SCIM and custom roles are paywalled in Pro Enterprise
- Fast product shifts toward agents may change what the paid tiers include
Pricing
View all features & details
Key Features
- Python pipelines run on swappable stacks
- Run, artifact and model version tracking with lineage
- Kitaru: agent trace import and replay
- Pro AI engineers for evals, reliability and cost
- Public agent benchmarks
Integrations
- MLflow, Arize Phoenix, Logfire
- pytest, promptfoo, DeepEval
- LangGraph, PydanticAI, OpenAI Agents SDK, Vercel AI SDK, Mastra
Real-World Usage
Named Customers
- AXA, JetBrains, ADEO, Leroy Merlin
- Airbus Defence and Space, Safran, IKEA, Vodafone, Rivian
Company
- ZenML GmbH, Munich
- Seed extension led by Point Nine (October 2023)
- Kitaru relaunched in 2026 for agent trace replay
How It Compares
| Feature | ZenML | Kubeflow | Metaflow | ClearML |
|---|---|---|---|---|
| Main job | Portable pipelines plus agent ops | Kubernetes ML platform | Python ML workflows | Tracking plus GPU scheduling |
| Runs on | Any stack | Kubernetes | AWS, Azure, GCP, Kubernetes | Hosted, VPC or on-prem |
| Agent features | Yes (Kitaru, Pro engineers) | Kubeflow MCP (2026) | No | GenAI App Engine |
| Paid entry | $39 per month | Free (managed by clouds) | Contact sales | $15 per user per month |
Sources: zenml.io · ZenML pricing · GitHub · ClearML pricing
Is this your tool? Add a badge