● Beta — KubeEra is in active development. Profiles are live; AI-perception measurement is rolling out.
KubeEra
M

MLRun

AI Infrastructure
Open source · Apache-2.0

MLRun is an open-source orchestration layer for building, running, and tracking ML pipelines on Kubernetes.

updated 2026-07-01

What it is

MLRun is an open-source orchestration layer for building, running, and tracking ML pipelines on Kubernetes. It handles the plumbing most teams end up hand-rolling: experiment tracking, feature store, serverless function packaging, batch/real-time inference serving, and pipeline orchestration (it can run standalone or sit on top of Kubeflow Pipelines / Argo). Think of it as the glue between a data scientist's notebook and a repeatable, versioned, auditable K8s workload.

Who builds it and why

MLRun originated at Iguazio, a real-time data-science platform vendor that was acquired by McKinsey (QuantumBlack) in 2023. The project lives under the LF AI & Data umbrella rather than CNCF. It's maintained by a mix of ex-Iguazio/McKinsey engineers and community contributors — 100 contributors on GitHub is a reasonably healthy number for a project this scoped, and commit activity through mid-2026 shows it's still actively developed, not abandoned. The motivation is straightforward: Iguazio's commercial platform needs an open orchestration core to drive adoption, and MLRun is that core.

Production readiness signal

Mixed. 1,678 GitHub stars is modest for a project that's been around several years and claims production MLOps use — for comparison, Kubeflow and MLflow both sit an order of magnitude higher. The current release tag, v1.12.0-rc16, is notable: shipping an rc16 as the "latest release" either means the stable 1.12.0 hasn't cut yet or release hygiene is loose. Sixteen release candidates before GA (or in place of GA) is a signal to pin versions carefully and test upgrades in staging, not assume drop-in stability. Apache-2.0 licensing and Python-first codebase are both fine — no vendor lock via license. CNCF maturity data isn't available because MLRun isn't a CNCF project; there's no independent graduation bar it's cleared. Treat production readiness as "vendor-backed open source," not "community-hardened standard."

Who should use this

Teams already running or evaluating Iguazio's commercial platform, where MLRun is the native orchestration layer — you get first-party support and tighter integration. Also a fit for K8s-native ML teams who want an integrated feature store + pipeline + serving story without stitching together five separate tools, and who are comfortable operating a moderately young project themselves. If your data scientists are already writing Python and you want git-trackable, containerized experiment runs without building that tooling in-house, MLRun covers a lot of ground in one package.

Who should NOT use this

  • Teams wanting a CNCF-governed, vendor-neutral standard. MLRun's roadmap is effectively controlled by one commercial entity (McKinsey/Iguazio). If multi-vendor governance matters to your procurement process, this isn't it.
  • Teams already committed to Kubeflow or SageMaker. MLRun overlaps heavily with Kubeflow Pipelines and cloud-native MLOps suites; running both adds redundant complexity for little gain.
  • Small teams without dedicated platform engineering. The feature store + serving + tracking stack is powerful but has real operational surface area. If you have one or two ML engineers, a lighter tool (MLflow for tracking, plain Argo for pipelines) will cost less to run.
  • Anyone needing rock-solid release stability today. Shipping against an rc-tagged "latest release" is a red flag for regulated or high-availability production inference paths until GA lands and stabilizes.

Alternatives

  • Kubeflow — broader CNCF-adjacent ecosystem, more community mindshare, heavier to operate.
  • MLflow — lighter-weight, best-in-class for experiment tracking, weaker on native K8s orchestration/feature store.
  • Metaflow / Flyte — code-first pipeline orchestration with strong versioning; Flyte in particular is more Kubernetes-native and CNCF-adjacent in spirit if not formally.

Pricing

MLRun itself is fully open source under Apache-2.0 — no license cost to self-host. Iguazio (McKinsey) sells a managed commercial platform built on top of it; that pricing is not publicly available and requires direct vendor contact.

Frequently asked

What is MLRun?+
MLRun is an open-source orchestration layer for building, running, and tracking ML pipelines on Kubernetes.
Who builds MLRun?+
MLRun originated at Iguazio, a real-time data-science platform vendor that was acquired by McKinsey (QuantumBlack) in 2023. The project lives under the LF AI & Data umbrella rather than CNCF.
Is MLRun production ready?+
Mixed. 1,678 GitHub stars is modest for a project that's been around several years and claims production MLOps use — for comparison, Kubeflow and MLflow both sit an order of magnitude higher. The current release tag, v1.12.0-rc16, is notable: shipping an rc16 as the "latest release" either means the stable 1.12.
Who should use MLRun?+
Teams already running or evaluating Iguazio's commercial platform, where MLRun is the native orchestration layer — you get first-party support and tighter integration.
Who should not use MLRun?+
Teams wanting a CNCF-governed, vendor-neutral standard. MLRun's roadmap is effectively controlled by one commercial entity (McKinsey/Iguazio). If multi-vendor governance matters to your procurement process, this isn't it. Teams already committed to Kubeflow or SageMaker.
What are the alternatives to MLRun?+
Kubeflow — broader CNCF-adjacent ecosystem, more community mindshare, heavier to operate. MLflow — lighter-weight, best-in-class for experiment tracking, weaker on native K8s orchestration/feature store.
How much does MLRun cost?+
MLRun itself is fully open source under Apache-2.0 — no license cost to self-host. Iguazio (McKinsey) sells a managed commercial platform built on top of it; that pricing is not publicly available and requires direct vendor contact.