OrlojHQ/orloj ? reverse-engineered prompt

Build me a full stack platform for running multi agent AI systems from YAML.

I want to be able to define agents, tools, models, memory, schedules, approvals, and policies in config, then have the system schedule tasks, let workers claim them, run the agent flows, route messages between agents, and record everything for later inspection. It should support local development in one process, but also be able to scale out with persistent state and message passing, so the same setup can move from a simple demo to something production ready.

Please include a command line tool, a REST API, a web UI, and Kubernetes friendly resources if that fits. I also want traces, logs, metrics, task history, and clear status for each run, plus guardrails like tool permissions and human approval steps. If you need to check current docs or best practices while building it, look them up online.