ClearML

A self-hostable MLOps platform that tracks experiments, versions datasets and automates ML jobs, with model serving backed by NVIDIA Triton.

ClearML is a self-hostable MLOps platform for researchers and engineering teams managing model development and production. It connects experiment records with datasets, training jobs and model deployment so teams can trace results back to the work that produced them.

Experiment tracking: Record code changes, package versions, parameters, metrics and model artifacts, including outputs from TensorBoard, Matplotlib and Seaborn.
Jobs and pipelines: Turn experiments into remote jobs, combine them into automated pipelines and run hyperparameter optimization across Kubernetes, cloud or bare-metal infrastructure.
Dataset versioning: Manage dataset versions using S3, Google Cloud Storage, Azure or NAS storage.
Model serving: Deploy model endpoints with monitoring and NVIDIA Triton support for GPU inference.
Remote development: Access JupyterLab and VSCode-server sessions on cloud or on-premises machines through the clearml-session module.

The Python SDK uses the Apache 2.0 license and connects to a ClearML Server that stores experiment, model and workflow data. The server is available as an open-source self-hosted deployment or a hosted service; ClearML Agent handles job orchestration and reproducibility.

ClearML also offers an enterprise infrastructure platform for on-premises, cloud and hybrid GPU clusters. Its controls include tenant isolation, role-based access and fractional GPU allocation. The enterprise GenAI App Engine deploys LLM APIs on those clusters with access control and monitoring; these capabilities are separate from the Apache 2.0 Python SDK.