cosmyx-zero/UniNet ? reverse-engineered prompt

Build me a Python app called UniNet that passively watches unidirectional IP traffic and flags possible cyber threats without ever sending anything back or inspecting payloads. It should take synthetic traffic or a PCAP file, turn flows into bursts and a graph view, then score them with a mix of rules, anomaly detection, and a graph based model, with sensible fallback behavior if the heavier ML extras are not installed.

I want a simple dashboard on localhost that shows alerts, clients, threat severity, and basic evidence for each finding, plus an API behind it. If possible, also include a clear explanation view for why something was flagged, and a read only helper that can answer questions about an alert from the stored evidence. Make it easy to run with one command, support optional auth, and include a demo mode, metrics, and tests so I can see it working right away.