androoAGI/starnet ? reverse-engineered prompt

Build me a local first desktop app where I can create a few AI agents, give each one a role, and watch them work inside a cute pixel art space station. The station should feel like the real app, not just decoration, so rooms, hallways, bays, and objects should reflect what the agents are actually allowed to do and what they are currently working on.

I want to bring my own API key or sign in with supported provider accounts, then launch multiple agents at once, each with its own workspace, memory, transcript, and permissions. Make it able to persist work on disk, show spend and run history, and let me open finished files from an outbox instead of digging through chat. If it makes sense, include basic support for schedules, reusable skills, and connected tools or connectors. Keep secrets on the machine, not in the UI, and make sure the app only shows state it can actually prove. If you need to, look up the current docs online and iterate from there.