Security
Pixels stay. Answers leave.
Your footage is often the most sensitive thing your company owns. The architecture treats it that way.
Perimeter holding · 6 rules enforced
rule 01
Pixels stay
Models run at the edge, next to your cameras. Your footage is processed on hardware you control, it is not shipped to us to be watched.
rule 02
Corrections travel, footage doesn't
When an operator fixes a detection, what gets recorded is the correction and the prediction it replaced, not the frame. That is what feeds retraining, so the model learns from your world without your world leaving the building.
rule 03
Answers leave
What crosses the perimeter is the result: detections, alerts, counts, delivered to Slack, email, or your own API. The evidence stays home.
rule 04
Edge or cloud, your call
Deploy on your own hardware or in your cloud. The loop is the same either way; the boundary is wherever you draw it.
rule 05
Learning without keeping
What the loop keeps is the review queue: the flagged evidence frames and the verdicts your experts made on them. Your streams stay on site, and the system gets smarter without them ever leaving.
rule 06
Models you own
Your detector is tuned to your cameras and downloadable to run on the edge, no cloud dependency, no per-image inference fee, and the weights are yours.
Running something sensitive?
Enterprise includes a security review - our engineers working through your infrastructure and your requirements before anything ships.
Prefer to see it first? Start with your footage