The funding announcement landed in early summer 2026—$35 million for Coram AI, a Sunnyvale startup trying to make physical security systems less reactive and more, well, intelligent. But the money, while substantial, wasn't really the story. What caught attention was the company's pivot away from searchable video archives toward something that sounds either prescient or overhyped, depending on whom you ask: autonomous security agents.
These agents, Coram says, don't wait around for someone to pose a question. They investigate incidents on their own.
It's an ambitious claim in an industry where "AI-powered" has become marketing wallpaper—slapped on everything from doorbell cameras to enterprise surveillance platforms. Yet Coram, now carrying $66 million in total funding from backers including Ansa Capital and Battery Ventures, insists this shift represents something more substantive than buzzword inflation. The company's new Deep Investigation capability is designed to correlate video footage, access logs, and visitor records automatically, reconstructing events across facilities without requiring security personnel to stitch together evidence frame by frame.
Whether that proves transformative or simply incrementally useful remains an open question. But the early deployment numbers suggest the market is at least willing to entertain the pitch.
When Cameras Do the Detective Work
Deep Investigation positions itself as what the company calls an "AI native research layer"—jargon that translates to a system capable of handling multi-step queries and producing shareable reports without constant human supervision. In practical terms, a security team can ask the platform to trace someone's movement through a building, cross-reference door access events, and surface anomalies automatically.
A forced door triggers video review. An unauthorized badge swipe pulls up associated camera feeds. A visitor who overstays their scheduled window generates an alert with full evidentiary context attached. The system, in theory, does the initial grunt work that traditionally eats up hours of a security analyst's day.
This falls under the umbrella of what the industry has started calling "agentic AI"—systems that operate semi-independently rather than waiting for prompts. It's a fashionable term, though the autonomy here is narrower than it sounds. Coram's agents automate correlation tasks; they don't make security decisions outright. Humans still review findings. What's changed, the company argues, is who carries the investigative load.
Still, there's something undeniably appealing about the idea of a system that notices patterns humans might miss, or simply don't have time to pursue. Whether it works as advertised at scale—across thousands of cameras, dozens of facilities, and the messy realities of physical security operations—is a different matter.
A Platform Play, Not Just Video Search
Coram has been building toward this since its 2022 founding, assembling what it describes as a unified platform that combines video security, access control, emergency management, and guest screening into a single dashboard. In physical security, where fragmented systems create blind spots and coordination headaches, consolidation carries real operational value.
The access control component, which launched in October 2025, works with standard Wiegand and OSDP readers—industry protocols that mean organizations don't need to rip out existing door hardware. Door events link directly to video feeds. The emergency management module handles panic buttons, lockdowns, and mass notifications, all tied to live camera streams. Guest management screens visitors against databases and tracks their movements through badge systems and video.
One example the company points to: tailgating detection that counts people entering through a door and cross-references the count against access card swipes. Discrepancies get flagged automatically, with relevant video surfaced for review. It's the kind of small but meaningful automation that adds up when you're managing security across multiple sites.
The architecture rests on Coram Point, a family of edge appliances that function as AI-powered network video recorders. These devices support anywhere from four to over 80 camera feeds depending on the model, with GPU memory reaching up to 20GB in the larger units. They're compatible with any ONVIF-standard IP camera, which allows organizations to repurpose existing infrastructure rather than starting from scratch.
Processing happens on-device instead of routing everything to the cloud—a design choice that reduces bandwidth consumption (the company claims roughly 30 kilobits per second per camera during idle viewing) and keeps sensitive footage local. Features like gun detection run entirely on the edge appliance, analyzing video for weapons without sending data over the internet. Coram frames this as both a privacy win and a performance advantage, though edge processing also means the intelligence of the system is constrained by what can run locally.
Traction in Schools, Healthcare, and Public Sector
As of mid-2026, Coram reported deployment across more than 1,500 sites in the U.S. and Canada. The customer roster includes Fortune 500 companies alongside names like 1-800-GOT-JUNK?, Hershey's Ice Cream, World YMCA, and Lakepointe Church. The company claims 4x revenue growth and triple the customer base since its $13.8 million Series A round in January 2025—metrics that sound impressive but are easier to achieve from a small initial base.
School districts and healthcare providers represent a significant chunk of deployments, sectors where security concerns intersect with regulatory scrutiny and limited IT budgets. Salem High School in Virginia piloted Coram's weapons detection system in 2025 at an annual license cost around $47,000, plus $40,000 in first-year equipment expenses, according to reporting from SecurityInfoWatch—pricing that reflected 2025 rates and may not represent current costs. Other implementations include a 26-school network operated by Inspire Development Centers, which migrated away from a proprietary ecosystem, and Point Defiance Zoo, which adopted the platform's AI-powered video search.
In February 2026, Coram secured a statewide cooperative purchasing contract in Indiana through ECPA/CIESC, allowing eligible public institutions to procure the platform without running separate competitive bids. It's a distribution win that streamlines adoption in the notoriously procurement-heavy public sector, though cooperative contracts don't guarantee actual sales—just easier access.
Autonomy Credentials from Self-Driving Cars
The founders' backgrounds add a layer of credibility to the autonomy narrative, even if physical security and autonomous vehicles occupy different worlds. CEO Ashesh Jain led autonomy efforts at Lyft Level 5 after stints at Zoox and a PhD from Cornell. CTO Peter Ondruška ran AI research at Lyft before moving to Toyota Woven following Toyota's acquisition of Lyft's self-driving unit in 2021. That pedigree shows in the platform's emphasis on computer vision and edge AI—core competencies in the autonomy world that translate reasonably well to security's video-heavy workflows.
The Series B was co-led by Ansa Capital and Battery Ventures, with participation from UP.Partners, 8VC, and Mosaic Ventures. Battery's Marcus Ryu, who joined Coram's board during the Series A, highlighted the autonomous investigation capability in the funding announcement as a key differentiator. Whether investors are betting on the technology or the team's ability to execute in a crowded market is probably some combination of both.
Coram operates in a competitive landscape that includes entrenched players like Genetec's Security Center, Motorola Solutions' Avigilon platform, the recently merged Brivo-Eagle Eye Networks entity, and Verkada's tightly integrated proprietary hardware ecosystem. Each offers some version of unified video and access control. What Coram emphasizes—hardware flexibility through ONVIF compatibility and the autonomous investigation layer built on transformer-based AI models—positions it as the platform for organizations that want intelligence without vendor lock-in.
Of course, flexibility can also mean complexity. Proprietary systems, for all their drawbacks, tend to "just work" out of the box. Open ecosystems require more integration effort. Which approach wins depends on the buyer's tolerance for customization versus their desire for simplicity.
The Autonomy Question
The shift from reactive search to proactive investigation represents a logical evolution in how physical security systems might operate. Whether "autonomous agents" becomes more than a marketing construct depends on how reliably the system surfaces actionable intelligence versus generating noise that still requires human filtering. Early deployment numbers—1,500-plus sites is meaningful traction for a four-year-old company—suggest receptivity. But the real test will be whether these agents genuinely reduce workload or simply redistribute it in different forms.
Coram claims SOC 2 Type II and HIPAA compliance, positioning itself for regulated industries, though those certifications are vendor-stated and should be verified by prospective customers during procurement. The company has grown to roughly 124 employees across offices in Sunnyvale, London, Bengaluru, Lehi, and Phoenix, according to LinkedIn data from around the time of the Series B.
As physical security converges with AI and edge computing—a convergence that feels inevitable even if the timeline remains uncertain—the central question isn't whether systems will get smarter. They will. The question is whether they'll get smarter in ways that matter to the people actually using them: security directors managing sprawling facilities, IT teams juggling integration headaches, compliance officers tracking audit trails.
Coram's bet is that autonomy in investigations is where the value lies, that the most pressing problem in physical security isn't capturing footage but making sense of it without drowning in data. The next year or two of enterprise deployments will reveal whether buyers agree—or whether this particular vision of autonomous security remains more compelling in pitch decks than in practice.
