Pieces, Not Stack
Most security programs already have plenty of technology in place. Edge devices watch doors, yards, parking lots, lobbies, and perimeters. A cloud platform handles video and access control. Analytics run inside cameras, NVRs, and software.
On paper the system looks sophisticated. In practice, incidents still bounce between systems and people. Detection happens in one place, investigation in another, and documentation often arrives days later in a report assembled from memory more than from evidence.
The problem usually isn’t capability. It’s architecture.
A real-time stack comes down to three connected ideas. The edge determines what happens in the first second. The cloud defines what’s true about the site, the schedule, and the history behind that alert. An agentic control layer in the cloud combines both and turns them into one coordinated response instead of a series of disconnected steps.
If you can’t sketch that relationship in a few strokes, you probably don’t have a stack. You have a collection.
Edge AI
The edge is where security meets reality.
It’s the camera on a pole, the mobile trailer in a yard, the robot moving along a patrol route, the intercom at a gate, the lights and speakers that can change what someone feels in the moment. Inside those devices, onboard analytics are increasingly capable of distinguishing people from shadows and vehicles from wind-blown debris.
The edge delivers three things nothing else can.
Presence, because the device is physically on site and can be seen and heard.
Speed, because it can detect, verify, and react in milliseconds without waiting for a round trip to a distant data center.
Resilience, because it can continue working even when network connectivity is slow or unreliable.
On its own, though, the edge is fast and narrow. It only sees what’s directly in front of its sensors. It doesn’t know who should be on site, whether a zone has had repeated incidents, or how a small event fits into a pattern across multiple locations.
That’s where the cloud comes in.
Cloud
The cloud acts as the system of record for the entire security program.
It’s where sites, cameras, access points, users, and policies come together. Video archives live there. Badge activity and door events are stored there. Device health, configuration, and integrations to IT, HR, and case management systems all reside there as well.
The cloud provides three things the edge cannot deliver alone.
Global visibility, allowing teams to see activity across many locations instead of one camera view.
Historical perspective, enabling analysis of weeks or months of events to identify patterns that a single device could never detect.
Control at scale, allowing administrators to update policies, manage devices, and connect security data to the rest of the organization.
But the cloud has its own limitations. It holds context and history, yet it sits one step removed from the physical scene. It relies on edge devices to carry out any real-world action.
To achieve real-time security, something must connect both worlds.
Agentic AI
Agentic AI operates in the cloud alongside video, access control, and device data. It functions as the control layer that interprets events and determines what should happen next.
Its role is to listen to events from devices and systems, interpret those signals in context, and then drive coordinated actions both at the edge and in the cloud.
First, it listens.
It receives detections from edge devices and events from other systems. It sees badge denials, armed states, schedules, and policies. It knows which sites are open or closed, which zones are sensitive, and which locations have had recent incidents.
Next, it reasons.
The system evaluates those signals within context. Is the motion occurring after hours or during a shift change? Is the individual expected to be on site? Does the pattern resemble a nuisance alert or a scenario requiring escalation?
Finally, it acts.
Agentic AI launches responses through both layers. It may trigger a live voice warning or visual deterrent from a nearby device while simultaneously opening an incident record, notifying stakeholders, and preparing a structured report.
The edge provides speed and presence. The cloud provides context and reach. Agentic AI connects both into one coordinated system.
One Incident
The easiest way to understand the stack is to watch it in motion.
Imagine someone climbing into a fenced yard after hours.
At the edge, a camera detects movement where no activity should occur. On-device analytics confirm that it’s a person rather than an animal or environmental motion. Within a fraction of a second, the device is ready to respond.
In the cloud, the system knows the site is closed. It sees that no access schedule allows workers to be in the yard and that a similar incident occurred in the same zone two weeks earlier.
Agentic AI receives both signals at once.
It identifies the pattern as an after-hours perimeter intrusion and selects an appropriate response. The nearest device issues a directed voice warning and activates lighting. At the same time, the monitoring team receives a concise notification with a live view rather than a generic alarm.
An incident record opens automatically and begins logging each action.
From the intruder’s perspective, the site responds immediately. From the operator’s perspective, the system has already handled the first steps and delivered enough context to guide escalation. From leadership’s perspective, the response is consistent, repeatable, and documented.
That is what a real-time stack looks like in action.
Real-Time Check
You don’t need to rebuild your entire environment to start moving toward this model.
Begin with a simple exercise: determine whether your environment behaves like a stack or a collection.
Ask which decisions occur at the edge without waiting for human input. If the answer is none, the devices on site are likely underutilized.
Look at what your cloud platform knows that a single site cannot. If the cloud behaves like a larger DVR rather than a contextual system of record, important capabilities are missing.
Then examine where intelligence currently stops. If alerts simply enter a queue waiting for manual review, there is no orchestration layer in place.
Finally, try pulling a complete incident record from detection to resolution. If it requires searching across clips, logs, messages, and notes, the system was not designed to produce evidence efficiently.
Real-time security begins with recognizing those gaps.
RAD In Practice
This architecture is not theoretical. It is how RAD structures its systems.
At the edge, RAD devices detect and respond locally. Fixed units monitor specific zones and deliver audio and visual deterrence. Mobile trailers provide coverage for remote or temporary sites. Autonomous patrol vehicles extend monitoring across larger areas.
Each device performs onboard detection, verification, and local response while maintaining reliable connectivity to the cloud.
In the cloud, organizations manage video, device status, and policy across multiple sites through their monitoring platforms. Those systems provide the history, visibility, and integration necessary to manage complex environments.
SARA Agentic AI operates between these layers.
She receives events from RAD devices, customer cameras, and third-party systems, evaluates those events in context, and determines what should happen next.
With SARA Edge, that intelligence can speak directly through devices on site. The edge system delivers the audio locally while SARA determines the language, tone, and response based on schedules, policies, and live conditions.
From the outside, the process looks simple. Something happens. The system verifies it, responds at the edge, informs the right people with clear context, and records the entire sequence automatically.
Underneath, it is a coordinated stack where edge devices and cloud systems operate together under an agentic control layer.
Closing Thought
Many organizations already own pieces of this future. They have intelligent edge devices and cloud-based management platforms.
What is often missing is the layer that connects those investments into a single system capable of detecting, deciding, and responding in real time.
The shift now is architectural. Instead of treating edge and cloud as separate purchases, organizations can design them as parts of one incident orchestration stack, connected by an agentic control layer.
Real-time security is no longer a claim. It is a design decision.
David Marsh
Vice President, Marketing
Robotic Assistance Devices
