Infrastructure & MLOps

Mlrun

MLRun is an open-source AI orchestration platform designed to build and manage continuous AI applications throughout their lifecycle. It supports data preparation, model training, deployment, and monitoring, integrating with development and CI/CD environments to automate production data workflows and machine learning pipelines. MLRun enables batch and real-time data processing, tracks data lineage, experiments, and metadata, and supports scalable resource management including containers and GPUs. It deploys real-time serving graphs using Nuclio for scalable inference and handles generative AI tasks such as retrieval-augmented generation (RAG), large language model evaluation, and fine-tuning. The platform provides a Function Hub with pre-built functions for ETL, auto-training with frameworks like Scikit-Learn and XGBoost, batch inference, and Azure AutoML. MLRun supports multi-cloud, on-premises, and hybrid environments, allowing collaboration across data, ML, software, and DevOps teams. It requires deploying several services including the MLRun API, UI, database, and Nuclio for full functionality, with Kubernetes preferred for backend setup.

Updated Dec 22, 2025open-source

Visit Mlrun ↗Visual Guide

Overview

MLRun is an open-source platform that automates end-to-end machine learning pipelines and real-time model serving with integrated lifecycle management.

Pricing

open-source

Continuous AI Application Development

Building and managing machine learning applications that require automated data preparation, model training, deployment, and monitoring.

Real-Time Model Inference

Deploying scalable real-time inference services using serverless functions integrated with Kubernetes or Docker.

Generative AI Tasks

Handling retrieval-augmented generation, large language model evaluation, and fine-tuning within AI workflows.

Quick Start

Install MLRun Client

Install the MLRun client package via pip and configure it for either a local or Kubernetes backend.

Import Function

Import a function from the Function Hub, such as 'hub://auto_trainer', to perform model training.

Create Project and Run Function

Create a project and run the imported function with inputs like datasets and parameters (e.g., model class, train-test split).

View Results

Use the MLRun UI to view results, metrics, and artifacts from your runs.

Deploy Serving Function

Deploy serving functions for real-time inference using the 'deploy_function()' method integrated with Nuclio.

📊

Strategic Context for Mlrun

Get weekly analysis on market dynamics, competitive positioning, and implementation ROI frameworks with AI Intelligence briefings.

Try Intelligence Free →

7 days free · No credit card

Assessment

Strengths

Automates machine learning pipelines, reducing engineering effort and eliminating boilerplate code.
Supports any IDE, framework, or third-party service across multi-cloud, on-premises, or hybrid environments.
Enables collaboration across data, ML, software, and DevOps teams through shared assets and metadata.
Automatically logs experiments, lineage, and results to support reproducibility and monitoring.
Scales training and serving workloads using built-in or custom functions on GPUs and containers.

Limitations

Requires deployment of multiple services including MLRun API, UI, database, and Nuclio for full functionality.
Backend setup is optimized for Kubernetes, with limited local deployment options.
Relies on integrations such as Nuclio for real-time model serving capabilities.