Tools¶
The tools used in this guide.
Core tools¶
- Code management: Git
- Package management: pip or uv as an alternative
- Data and model versioning: DVC
- ML experiment reporting: CML
- Pipeline orchestration: GitHub Actions
- Cloud infrastructure: Google Cloud
- Model packaging and serving: BentoML and Docker
- Model deployment: Kubernetes
- Model observability and monitoring: Fluent Bit and Evidently AI
- Data annotation: Label Studio
The following chapters explain each tool in detail.
Related tools¶
This guide covers one toolset, but alternatives exist for every stage. For a broader compilation, see MLOps.toys.
Data management¶
Alternatives to DVC.
- LakeFS - Git-like version control for data lakes
- DagsHub - Data science collaboration platform
- DoltHub - Collaborative versioned databases
- Delta Lake - Open-source storage layer for lakehouses
Experiment tracking¶
Alternatives to CML for tracking experiments and visualizing metrics. CML reports results inside CI/CD pipelines, these tools track and visualize experiments instead.
- Guild AI - Open-source toolkit for running, tracking, and optimizing ML experiments
- Aim - Open-source, self-hosted tool for tracking and visualizing ML experiments
- TensorBoard - Open-source toolkit for visualizing ML experiment metrics and model graphs
Model monitoring¶
These are alternatives to Evidently AI for monitoring models in production.
- NannyML - Detect model and data drift, including estimated performance degradation, without ground truth labels
- Deepchecks - Test and validate ML models and data, with a library or self-hosted UI
- Seldon Alibi Detect - Algorithms for outlier, adversarial, and drift detection
Logging and observability¶
Alternatives to Fluent Bit for collecting, processing, and forwarding logs and observability data.
- Vector - High-performance, end-to-end observability data pipeline for logs, metrics, and traces
Data annotation¶
Label Studio handles many data types, but most competitors specialize in one.
See the
awesome-data-labeling
repository for specific alternatives.
Pipeline orchestration¶
Alternatives to GitHub Actions.
- GitLab CI - DevOps platform with built-in CI/CD and container registry
- Gitea - Self-hosted Git service with built-in CI/CD using GitHub Actions-compatible syntax
- Forgejo - Self-hosted Git service and soft fork of Gitea with GitHub Actions-compatible workflows
Model packaging and serving¶
Alternatives to BentoML for packaging and serving models.
- MLEM - Open-source tool to simplify ML model deployments
- Cog - Package machine learning models in standard, production-ready containers
- Seldon Core - Open-source platform to deploy ML models on Kubernetes
- Kubeflow - ML workflows on Kubernetes, including training and serving
Container tools¶
Alternatives to Docker.
- Podman - Daemonless, open-source tool for running, building, and sharing OCI containers and images
Self-hosted infrastructure¶
Tools for running the MLOps stack on your own hardware instead of managed cloud services.
- CNCF Landscape - Graduated CNCF projects considered production-ready
- Kubespray - Deploy production-ready Kubernetes clusters on bare-metal or virtual machines
- Argo - Kubernetes-native continuous delivery (Argo CD) and workflow orchestration (Argo Workflows)
- Harbor - Self-hosted container registry with vulnerability scanning and RBAC
- Distribution Registry - Lightweight local container registry, also known as Docker Registry
- Helm - Package manager for Kubernetes; commonly used to install and manage Argo, registries, and CI runners
- Docker Swarm - Simpler, built-in container orchestration alternative to Kubernetes
End-to-end platforms¶
Tools that cover the whole ML lifecycle in one platform. They are often opinionated, so this guide prefers composable tools.