We built LiteLLM Operator because running LiteLLM on Kubernetes deserved more than a manual Helm workflow. LiteLLM already gives teams a unified way to route across 100+ LLMs through an OpenAI-compatible interface. Our goal was to make operating that gateway on Kubernetes feel native: declarative CRDs, reconciliation, production-ready defaults, and a cleaner path for managing models, users, teams, and access in real environments. A big part of the value is closing the gap between platform engineering and day-to-day gateway administration. Instead of treating deployment and management as separate worlds, LiteLLM Operator is designed to make Kubernetes the control plane for LiteLLM operations. That means less drift, less glue code, and a workflow that fits how modern platform teams already ship software. We’d love feedback from platform engineers, MLOps teams, and anyone running LiteLLM in production: what part of operating AI gateways is still more painful than it should be?

LiteLLM Operator

Run LiteLLM on Kubernetes the operator way

Run LiteLLM on Kubernetes the operator way