berriai/litellm

LiteLLM is an open-source Python library and proxy server that gives you a single, consistent way to talk to over one hundred different AI language model providers, including OpenAI, Anthropic, Google Gemini, Amazon Bedrock, Azure, and many others, without having to learn each provider's unique API format. The problem it solves is fragmentation. Every AI provider has its own SDK, its own authentication method, its own request and response shapes, and its own error messages. If you build an application using OpenAI's API and later want to switch to Anthropic, or run multiple models side by side, you would normally have to rewrite significant portions of your code. LiteLLM removes that friction by translating all provider APIs into the single OpenAI format, the industry's closest thing to a standard. It works in two modes. As a Python SDK, you import it directly into your code and call any model through the same completion() function regardless of provider. As an AI Gateway (a proxy server you deploy yourself), it acts as a centralized intermediary that your entire team routes requests through. The gateway adds production-ready features: virtual API keys (so team members never see the real provider credentials), cost and usage tracking per key, rate limiting, load balancing across providers, and guardrails that filter harmful content. A team would use LiteLLM when they want to experiment with different AI models without rewriting their application each time, when they need visibility into how much they are spending on AI calls, or when they want to give developers safe access to AI models without exposing raw API keys. The tech stack is Python. The proxy server is a FastAPI-based HTTP service that can be self-hosted with Docker or deployed to cloud platforms.