omnilabs-ai/OmniRouter ? reverse-engineered prompt

Build me a Python app called OmniRouter that gives me one simple API for a bunch of modern LLMs, so I can switch models without changing my app each time. I want it to feel like a clean OpenAI style chat and image API, with a router that can send requests to the best or cheapest model based on simple rules. Please include the reasoning API too, with streaming support and an easy way to control how much reasoning a model uses.

Make sure the server side is organized cleanly, and that there are basic docs and a small test setup I can run to check chat and image requests locally. Keep it easy to set up with a virtual environment and a requirements file, and if you need current docs for any provider or FastAPI details, look them up online. I’d also like the code to be easy to extend later for more providers and routing rules.