LiteLLM is an open-source model gateway and Python SDK that provides a unified interface for calling many different LLM, embedding, image, audio, reranking, and agent providers. It translates provider-specific APIs into common formats, including OpenAI-compatible endpoints, so applications can switch models or vendors without rewriting large parts of their integration code.

The LiteLLM proxy adds centralized routing, virtual keys, authentication, budgets, rate limits, usage tracking, fallbacks, logging, and observability for teams operating multiple model providers. It also supports newer agent and MCP-related interfaces, making it useful as an infrastructure layer between applications and a changing AI model ecosystem. LiteLLM is particularly valuable for developers and platform teams that want provider flexibility, cost governance, and a single gateway for production AI traffic.