awaisshahid965/sage ? reverse-engineered prompt

Build me a small AI support chat app called Sage for a fictional gadget store called Pebble.

I want a web chat experience where a user can ask product and shipping questions, get one reply at a time or see the answer stream in real time, and keep chatting in the same conversation instead of starting over each time. Please make it so the assistant can work with swappable model backends, and keep the conversation on the server so the client only needs to hold a conversation id.

There should also be a way to inspect what the model is thinking at the start of an answer, especially the top token choices and their probabilities, because I want to see how confident it is. If it helps, include a simple command or page for that.

Use a clean FastAPI style backend, a simple chat UI, and make the whole thing easy to run locally with sensible defaults and good tests. If you need to look up current docs online, go ahead.