OpenHands/litellm — reverse-engineered prompt

Build me a Python based LLM gateway and SDK that gives me one simple OpenAI style way to talk to lots of different model providers. I want to be able to switch between providers like OpenAI, Anthropic, Azure, Bedrock, Vertex, Groq and others just by changing the model name, instead of rewriting my app every time. It should work two ways, as a Python library I can import, and as a local proxy server I can run and point existing OpenAI clients at with a base URL.

Please make the local setup easy, with environment variable based keys, a sample config, Docker support, and a quick start that lets me send a chat request right away. It should cover the common endpoints people expect, like chat, embeddings, images, audio, and batch style calls if supported. If it fits the project, include the gateway pieces for virtual keys, MCP tool access, and agent to agent calls too.

Keep the developer experience simple, with clear docs, examples, and sensible defaults. You can look up current docs online if you need to.