Senzorium Fusion takes the streams from every edge node and folds them into a single multi-modal frame — flow, behavior, and system signals correlated per host, per tenant, per second. This is where perception becomes decisive.
snapshot · refreshes every 60 seconds
Each modality on its own is noise. Fusion turns them into signal.
Connection-level features: 5-tuples, byte rates, entropy, protocol fingerprints, port scan patterns.
Per-host and per-tenant baselines: typical traffic shape, periodicity, peer-set drift, departure-from-normal scoring.
Edge node telemetry: auth events, file integrity, process lineage, kernel events — all aligned to the same timeline.
Fusion's output is a single perceptual frame per (host, second) tuple. It carries the modal verdict from each lane plus a meta-verdict that weights them by recent accuracy. Downstream engines reason on the frame, not on raw signals — which is why detection latency stays low and false positives stay flat.
The live demo shows the three modalities collapsing into a single frame.
