labsiqbal/proxima ? reverse-engineered prompt

Build me a self hosted web app that acts like a control room for the AI agents I already use.

I want to open it in a browser on my laptop or phone, sign in once as the owner, and then use it to chat with an agent, turn that chat into a reviewable plan, and run jobs that I can inspect before they are marked done. It should support delegating work to different agent CLIs I already own, like Claude Code or Codex, without shipping any model or credentials itself.

Please include a main chat area, a place to manage tasks and reusable recipes, a history or archive for finished work, and a way to review file changes before merging them back. If it makes sense, add a master mode for handing off bigger outcomes to multiple workers, plus a simple artifact viewer for things like docs, diagrams, and feedback. Keep it self hosted, practical, and honest about what is running and what still needs my attention.