"What Happened in the Parking Lot Yesterday?" — Analyzed, Not Just Played Back

Rhombus + Claude Series

An incident happened in your parking lot sometime between 10 PM and 6 AM. You don’t know when. You don’t know what. You just need someone to figure it out.

In a traditional camera system, that means someone is about to spend four hours of their day watching eight hours of footage at 2x speed. They’ll miss things. They’ll get tired. They’ll make a call on what “unusual” looks like, and they’ll be working from memory by the time they finish. It’s the exact kind of work AI was invented to eliminate — and yet, it’s still the default workflow in most security operations centers.

The problem isn’t that AI can’t analyze video. The problem is that most camera platforms don’t expose their footage in a form AI can use.

Rhombus does. Any camera’s footage is accessible as structured, timestamped frames — ready to hand to a model.

For the developers:

With Claude and the Rhombus plugin, reviewing an overnight window is one prompt:

“Summarize everything that happened in the east parking lot between 10pm and 6am. Flag anything unusual.”

Claude pulls the activity frames from Rhombus, analyzes them, and returns a written timeline: “3 vehicles entered between 11pm and midnight. One unrecognized vehicle lingered near the south entrance at 3:47 AM for 12 minutes.”

The payoff: Footage stops being something humans watch and starts being something AI reviews. Hours of manual scrubbing become seconds of analysis. And for the first time, reviewing every overnight window at every site becomes feasible — because the bottleneck was never the cameras. It was the time it took to look at them.