Walk into any large facility today—a hospital, a warehouse or a school—and chances are the walls are lined with cameras. Hundreds of them, maybe thousands, recording around the clock. But even though most of those cameras capture and store data, what they cannot do is tell you anything useful at the moment. When something goes wrong, someone then uses the system and conducts an investigation. By then, it's already too late.

That disconnect between seeing and understanding is what Lumana was built to close. The company's latest innovation, VIA-1, is the engine behind that ambition: a proprietary video intelligence model the company describes as the "world's first continuous-learning AI built for IP cameras," one that adapts to each camera's environment rather than applying the same fixed, rule-based AI across the board.

"Think of it like the difference between asking a general question with no context versus working with a system that has deep, site-specific knowledge built up over time," says Jordan Shou, VIce President of Marketing at Lumana. "The result is a solution that becomes smarter, more accurate and more predictive the longer it's deployed—rather than degrading or stagnating, as static video security and analytics systems typically do."

The Camera That Keeps Learning

When a camera connects to the Lumana platform, VIA-1 deploys a dedicated model instance to that specific device. From that point, the model starts building a picture of what "normal" looks like in that exact environment—adjusting for lighting shifts, activity patterns, seasonal changes and the kinds of situational quirks that vary from a hospital lobby at noon to a loading dock at 3 a.m.

This is a meaningful departure from how most video security AI works. Traditional systems are trained once and frozen at deployment. They can flag motion and identify people or vehicles, but they cannot account for context and they generate false alerts at a rate that slowly trains security teams to stop paying attention. Shou claims VIA-1 reduces false alerts by up to 90% compared to legacy providers. In one deployment, Shou says a continuous-learning system maintained twice the detection accuracy of a static model through seasonal changes, while the static model's performance dropped noticeably during winter months.

The market Lumana is targeting reflects how overdue this shift has been. The video analytics industry is projected to grow from roughly $12 billion today to nearly $38 billion by 2030, driven in part by the move away from fixed, rule-based systems that required constant manual adjustment and updates.

How It Works

VIA-1 is camera-agnostic by design. The platform runs with hardware from Hanwha, Pelco, Avigilon, Axis, Bosch, Vivotek and others, requiring no replacement of existing infrastructure. It also connects to access control systems, IoT sensors and other enterprise platforms. Lumana currently processes analytics on billions of images per day.

The platform is built around four layers: aggregation, AI models, AI agents and application. VIA-1 operates as the intelligence layer and backbone of the system and its output activates four AI agents—Monitor, Response, Investigate and Insight—each handling a different part of the video security workflow. Monitor sends real-time alerts for detected activity. Response can automatically notify staff, trigger loudspeakers, initiate lockdowns or dispatch emergency services. Investigate lets teams search footage in seconds using natural language or granular parameters. Insight surfaces operational data from video—occupancy, dwell time, queue length, heatmaps—through visual dashboards.

"Everything is managed from an intuitive, single pane of glass to make our AI actually usable for enterprise organizations," Shou explains.

Results in the Field

At Gateway Community & Technical College in Kentucky, Lumana's platform monitors the campus around the clock, alerting staff to weapons, unauthorized access and other incidents without requiring someone to watch every feed. In retail, the company notes that one multi-location operator with over 200 cameras from different manufacturers consolidated everything into a single view, cutting investigation time from hours to seconds and picked up operational uses along the way, including customer traffic patterns, dwell time and cash room access monitoring.

The numbers from large-scale deployments tell their own story: at a major software conference with more than 50,000 attendees across multiple venues, Lumana says it processed 3.4 million detections at 99.99% object accuracy. In a city during an annual holiday event with a strict 16,000-person venue capacity and nearly 100,000 daily visitors, the system tracked all 14 entrances over three days and logged more than 160,000 entries and exits at 99.5% counting accuracy.

Other customers include McDonald's, Meta, the Minnesota Twins Baseball Club, University of Miami and Tuff Shed. VIA-1 is embedded in every deployment from day one—no separate licensing, no add-on—so its AI capabilities and the performance their customers receive improve with every new camera connected to the platform.

50,000 Cameras and Counting

Lumana surpassed 50,000 cameras on the platform in under two years, a pace the company says makes it one of the fastest-growing AI video security companies in the world. In February, it closed a $40 million Series A led by Wing Venture Capital, with continued backing from Norwest and S Capital, bringing its total funding since inception to $64 million. The company says that capital is going toward accelerating go-to-market efforts and continued development of VIA-1 and AI agents.

The global AI video surveillance market, Lumana projects, will more than double over the next several years. The company is betting VIA-1 puts it at the center of that growth—not by adding intelligence to cameras, but by making the cameras themselves intelligent, enabling them to monitor, investigate and act automatically in the physical world and at enterprise scale.