dabster108/nexus-mac-agent ? reverse-engineered prompt

Build me the full NEXUS app, a local AI assistant for macOS that runs on my machine and helps me do things safely.

I want a browser based chat interface, a local backend, and a Mac helper process that the model can talk to only through approved tools. It should answer questions about my machine, remember important facts across sessions, notice changes on my Mac, and suggest useful actions, but never do anything risky without asking me first. Safe read only actions should run right away, anything that changes something should require confirmation, and blocked actions should never run.

Make it feel like one coherent product, with live task updates, a clear trace of what happened, and a way to explain actions in plain English. It should start locally on macOS, bind only to localhost, use the model provider already expected by the project, and keep the architecture secure and permission aware. If you need current docs for any library, look them up online.