Jettysnigdhan/AdaptiveAI- ? reverse-engineered prompt

Build me an OpenAI compatible LLM routing gateway that I can point Cursor or VS Code at and use like a normal chat API.

I want it to accept requests at /v1/chat/completions, route each prompt to the right model automatically, and keep track of quality, latency, and cost. It should start with a small model when possible, then escalate to bigger models if the answer looks weak, and it should also support a simple rule based mode and a fixed model mode for comparison. Please make it work with a few providers, including cloud and local options, and let me switch providers from an env file.

I also want a live dashboard that shows request logs, route split, cache hits, latency, and quality over time. Add clear telemetry headers on every response so I can see which model was used and whether escalation happened. Include the basic local setup, tests, and benchmark runner so I can reproduce the results. If anything needs current docs, look them up online.