bleetcoding/nvidia-floating-chat ? reverse-engineered prompt

Build me an Android app that works like a floating AI chat bubble, similar to a chat head, with a persistent overlay you can drag around and tap to open the full chat. It should be an OpenAI compatible client, with the NVIDIA endpoint already filled in as the default, but the user can change the API key, endpoint, and model themselves. Keep conversations on the device, support starting new chats, streaming responses as they arrive, cancelling a response, copying and selecting text, editing a message and sending it again, and regenerating replies.

Also let users attach images or text files from the system picker and only send those files with the message they chose. Make the setup feel simple, with a connection test and model discovery if possible. Use secure local storage for the key, and make the floating bubble work with the Android overlay permission and a foreground service notification. Look up current docs online if you need to.