Seldon Core is an open-source MLOps and LLMOps framework for deploying, managing and scaling machine learning models and modular AI applications on Kubernetes. Developed by Seldon, the GitHub project has just under 4,800 stars and, in its Core 2 generation, has evolved from single-model serving into a platform for composing data-centric AI systems from models, pipelines and infrastructure components.
Core 2 deploys models in a standardized way across a wide range of frameworks, on-premises or in any cloud, with production concerns built in rather than bolted on: Kafka-based data pipelines connect components into modular applications, an overcommit feature lets you deploy more models than memory would normally allow, and experiment tooling routes traffic between candidate models for A/B tests and shadow deployments. A scheduler and dedicated data plane keep thousands of models manageable in one cluster. Core 2 was a ground-up redesign informed by years of running the original Seldon Core in production at banks, telcos and public sector organizations, with an emphasis on composability: models, feature transformers and explainers are deployed as reusable pieces of larger AI applications.
Freemium. Seldon Core is open source and free to self-host; Seldon sells commercial products and enterprise support around the platform for organizations that need them.
Platform and ML engineering teams running serious model fleets on Kubernetes who need standardized deployment, experimentation and cost control across many production models.