The complete MLOps tools guide with curated recommendations for every category and a starter stack.
Experiment Tracking: MLflow
Feature Store: Feast
Pipeline: ZenML (or Airflow)
Model Serving: BentoML
Monitoring: Evidently
Data Versioning: DVC
Vector DB: pgvector
LLM Framework: LangChain
Containers: Docker
This covers 80% of MLOps needs.
| Tool | Pricing | Best Feature | Pick? |
|---|---|---|---|
| MLflow | Free (OSS) | Industry standard | START HERE |
| W&B | Free tier+ | Best UI | Best UX |
| Comet | Free tier+ | Code tracking | Good alt |
| Tool | Best For | Curve | Pick? |
|---|---|---|---|
| ZenML | ML-first pipelines | Low | Best for ML |
| Kubeflow | K8s-native | High | K8s shops |
| Airflow | General | Med | You know it |
| Prefect | Modern Python | Low | Airflow alt |
| Tool | Best For | GPU? |
|---|---|---|
| BentoML | General serving | Yes |
| vLLM | LLM serving | Required |
| Triton | Multi-framework | Optimized |
| FastAPI | Simple custom | Manual |
| Tool | Focus | Pick? |
|---|---|---|
| Evidently | Drift, quality | START HERE |
| Langfuse | LLM observability | For LLMs |
| Arize | Full ML observability | Production |
Solo/Small: MLflow + ZenML + BentoML + Evidently + Docker
Mid-size: Add W&B, Feast, Kubernetes, CI/CD
Enterprise: Cloud-native (SageMaker/Vertex/Azure ML) + K8s
Start simple. Add complexity only when you feel the pain.