BerriAI/litellm ? reverse-engineered prompt

I want to build a universal translator for LLM APIs. It should be a Python library that lets me call over 100 different models from providers like OpenAI, Anthropic, Google, and Cohere using the exact same code and the standard OpenAI format.

More importantly, build this out as a full-fledged proxy server or AI gateway. I want to be able to run this server, maybe in Docker, and then just point all my apps to it. The gateway should accept standard OpenAI API calls and translate them for whatever model I've specified—whether it's GPT-4o, a Claude model, or something from Bedrock. It needs to handle things like API key management, tracking costs across all the different providers, logging requests, and maybe even supporting features like load balancing. The main goal is to have one consistent endpoint to manage all my LLM usage without having to change my app's code.