evermind-ai/raven ? reverse-engineered prompt

Build me Raven, a multi agent workspace where I can give one task and it breaks the work into a clear workflow, then routes pieces to specialist helpers for research, coding, design, and long running unattended jobs.

I want one simple surface where I can see the plan, the agents working, their outputs, and the final result in a clean way. It should keep context between sessions, let me pick or coordinate multiple agents for the same task, and make it easy to follow what happened step by step. If you need to look up current docs online while wiring it up, go ahead.

Please include a usable web interface and a terminal style interface, basic support for running demo workflows, and a way to connect to different AI providers cleanly. Make it feel like a real product I could use to delegate complex work, not just a toy chat app.