Xeusee pulls the sub-streams from your existing network video recorder — no new cameras, no rewiring. Each stream gets motion gating first, so the detector only spends time on frames that changed. In a real building, where most cameras are quiet most of the night, that is what lets one box keep up with forty.
The engine is written in Rust and runs detection, tracking and rules on the box. Events, snapshots and health reports go out over an outbound-only connection; your video never has to leave the building to be analysed.
Your cameras are already recording. Now they can notice.
Xeusee · video AI for buildings
Xeusee is a small analysis box that sits next to your existing NVR, watches up to 40 cameras in real time and raises an event when something actually happens — intrusion after hours, loitering, a line crossed. It only ever dials out.
Apply for a pilot
Measured, not promised
Numbers from our own test bench: one Apple M1 Max box, 40 live 704×576 streams.
videocam
40
Cameras per box
Ingested at 5 fps each with zero dropped frames
speed
70
Frames analysed / second
Worst case, every camera in motion at once
call_made
0
Inbound ports
The box only connects out. Nothing to open on your firewall.
gavel
100%
Commercially licensed models
A license gate refuses anything AGPL or GPL
Plugs into the NVR you already own
Fits the building you have
- check_circleWorks with existing NVR sub-streams
- check_circleMotion gating before detection
- check_circleDetection, tracking and rules on the box
- check_circleOutbound-only connection, no port forwarding
Built to be trusted on the night it matters
Scale, proof and privacy — designed in, not bolted on.
Forty cameras, one box
On our bench a single Apple-silicon box ingests 40 streams at 5 fps with zero dropped frames, with a 4.3 ms-per-frame detector on the Neural Engine. When the box is saturated it marks frames as unobserved instead of pretending.
Signed evidence packages
Every event ships with its frames’ SHA-256, the model’s SHA-256, the rules version and the event record, signed with Ed25519. Anyone can re-verify later that nothing was edited.
Behaviour, not faces
Xeusee watches what happens — someone in a closed area, a person lingering at a door, an object left behind — without building face profiles of residents or visitors.
How a pilot runs
Careful by default: shadow first, alerts second.
1 · Site facts
A short questionnaire about your NVR, cameras and the risks you care about.
2 · Zones & rules
We start with 8–10 cameras: draw zones, set after-hours intrusion and loitering rules.
3 · Shadow, then live
Two weeks of shadow mode that only records events, threshold tuning, then live P1 and P2 alerts.
FAQ
help_outlineDo I need new cameras?
help_outlineDoes video leave the building?
help_outlineWhat can it detect today?
help_outlineWhich hardware does it run on?
help_outlineWhat does a pilot cost?
Apply for a Xeusee pilot
Tell us about your building: number of cameras, your NVR brand, and what keeps your security team up at night.
Forty cameras. One box. Zero open ports.
Video AI that fits the building you already have.