sydasif/litellm-proxy ? reverse-engineered prompt

Build me a Docker based AI proxy gateway that lets Claude Code talk to several model providers through one local endpoint.

I want it to route requests through LiteLLM, support multiple backends like NVIDIA NIM, OpenCode Zen, Agnes AI, and Gemini, and handle load balancing, failover, and basic rate limits so one provider does not get hammered. It should expose OpenAI style chat completions and work smoothly with Claude Code by pointing it at the proxy URL.

Please also include virtual API keys with budgets and model access limits, a simple admin UI to manage keys and see usage, plus health check endpoints so I can tell when it is alive and ready. Add Redis caching and Postgres for persistence if that is part of the setup. Make the whole thing easy to run with docker compose, include a sample env file, and set it up so I can paste in my provider keys and get started quickly.