hugalafutro/llm-convo — reverse-engineered prompt

Build me a small web app where I can make two AI chat backends talk to each other. Each side should have fields for a base URL, optional API key, character name, optional system prompt, and a Connect button that checks the backend and then lets me pick from the models it offers. I want to type an opening message, or have either side generate an in character starter, then run a conversation for a set number of rounds, like 1 to 30, with both replies streaming live into chat bubbles.

Please show any hidden thinking in a collapsible section if the backend sends it, and add simple stats like time, model, token count, and speed. Include Stop, Clear, dark and light mode, friendly connection toasts, and remember my settings and model choices in the browser. Keep the transcript nicely formatted with safe markdown so random output cannot inject scripts. It should run locally in Python and also be easy to launch with Docker. Look up current docs online if you need to.