MLSysOps Framework

The MLSysOps Framework is the open‑source outcome of the EU‑funded MLSysOps Horizon Europe project (Grant ID 101092912), running from Jan 2023 to Dec 2025. Its aim is to deliver an AI‑enabled, agent‑based platform for autonomic, cross‑layer management of compute, storage, and network resources across cloud, edge, and IoT environments.

Key Objectives

• Provide an open, AI‑ready framework for scalable, trustworthy, explainable system operation across heterogeneous infrastructures.
• Enable continual ML learning and retraining during runtime via hierarchical agents.
• Support portable, efficient execution using container innovation and modular, FaaS-inspired offloading.
• Promote green, resource‑efficient, and secure operations while maintaining QoS/QoE targets.
• Facilitate realistic evaluation using real-world deployments in smart‑city and precision‑agriculture scenarios.

Core Components

• Hierarchical Agent Architecture: Interfaces with orchestration/control systems and exposes an ML‑model API for plug‑and‑play explainable/re-trainable models.

• Telemetry & Control Knobs: Collects metrics across the continuum and adjusts configuration (e.g., compute, network, storage, accelerator usage) dynamically.

• Distributed FaaS‑style Executor: Enables function offloading across tiers to optimize latency, energy, and performance.

• Explainable ML & Reinforcement Learning Module: Offers transparent decisions, highlighting input factors influencing agent actions.

• Use-cases: Includes real applications focusing on smart cities and agriculture.

Repository Contents

Directory	Description
agents/	Core autonomic agents with policy-based plugins and ML/analytics
orchestrators/	Scripts to facilitate testbed setup
mlsysops-cli/	Tool to manage MLSysOps-related descriptors (agents, applications, etc.)
northbound-api/	Glue API from the CLI to the core Agent framework
docs/	Internal and public-facing documentation

Getting Started

Prerequisites

• Kubernetes v1.26+
• kubectl, karmada
• Python 3.10+
• Access to a 4-node testbed environment

Quick Start

Install the CLI tool:

pip install mlsysops-cli

Given an ansible inventory to setup 4 nodes in inv.yml, you can deploy the framework:

mls framework deploy-all --inventory inv.yml

Create and deploy an example application:

mls framework create-app-test-description
mls apps deploy-app --path mlsysops-app-test-description.yaml

See docs/ for detailed component setup guides.

Documentation

Check the full documentation at docs.mlsysops.eu

Contributing

We welcome contributions from the community!

Browse good first issues

Review our CONTRIBUTING.md

Follow our CODE_OF_CONDUCT.md

License

This project is licensed under the Apache 2.0 License.

Acknowledgements

This framework is developed as part of the Horizon Europe MLSysOps Project (Grant ID 101092912), coordinated by the University of Thessaly, with contributions from 12 European partners across academia and industry.

Learn more at mlsysops.eu