Seldon

Seldon is an enterprise MLOps platform built around Kubernetes for serving, scaling, and monitoring machine learning models in production. Its flagship pieces - Seldon Core 2, MLServer, and the Alibi explainability and drift-detection libraries - let teams package models as microservices, chain them into real-time inference pipelines (with native Kafka integration), and run canary or A/B rollouts with traffic shifting between model versions. It handles both classical ML and modern LLM/generative AI workloads on the same substrate.

The target buyer is a platform team or Chief AI Officer trying to standardize how dozens or hundreds of models get deployed across EKS, AKS, GKE, or on-prem clusters. Multi-model serving with memory overcommit is a real differentiator for cost - you can pack many small models onto shared replicas instead of one-pod-per-model. The open-source core is free; the enterprise tier (now bundled under TrueFoundry following their combination) is quote-based and aimed at regulated industries that need audit trails, RBAC, and governance.

Seldon plugs into the usual MLOps stack - MLflow, Hugging Face, Prometheus/Grafana, Istio, LangSmith - and exposes both REST and gRPC inference endpoints. It is decidedly not a no-code product: expect to write Kubernetes manifests and understand service meshes to get value from it.