Ambient.ai vs Spot AI vs Eagle Eye: AI-Native VMS 2026

Updated October 2026: Ambient.ai's Pulsar reasoning model and its October 2026 Series C, Spot AI's move to camera-agnostic “AI coworkers” sold only through partners, and the Brivo Eagle Eye analytics line-up after the merger.

Why this matters

If you are choosing a Video Management System (the software platform that ingests, records, and manages many camera streams, abbreviated VMS), the established platforms profiled earlier in this block — Milestone XProtect, Genetec Security Center, and Avigilon under Motorola Solutions — are only half the market. A newer group of vendors rebuilt the VMS around two ideas the incumbents added later: artificial intelligence as the point of the system, not a feature, and the cloud as the default home, not an option. This profile is for a security integrator, product manager, retail or operations lead, or enterprise security lead deciding whether an analytics-first, cloud-first platform fits — and what it really trades away. Read it to understand what "AI-native" buys you, how Eagle Eye Networks, Spot AI, and Ambient.ai differ from each other, and where each one fits before you put it on a shortlist.

Ambient.ai vs Spot AI vs Eagle Eye Networks: the short answer

Ambient.ai, Spot AI and Eagle Eye Networks are not three versions of one product; they differ by what they replace. Ambient.ai is a threat-detection layer you add on top of the cameras and the VMS you keep. Spot AI is camera-agnostic edge AI: an on-site appliance reads your existing cameras and runs AI agents that act, not just alert. Eagle Eye Networks, part of Brivo since December 2025, is a full cloud VMS that becomes your system of record. As of October 2026, all three work with the cameras you already own.

• Ambient.ai vs Eagle Eye Networks for enterprise security: choose Ambient.ai if you are keeping your Genetec or Milestone VMS and want proactive alerts on tailgating, loitering, weapons and perimeter breaches with no facial recognition; choose Eagle Eye if you want to replace on-site recorders with a cloud VMS across many sites. The two can in principle run side by side, because Ambient.ai sits on top of a VMS rather than replacing one; confirm the Eagle Eye integration with both vendors first.
• Spot AI vs Ambient.ai for facility security and video analytics: Spot AI fits mid-size, multi-location operators (manufacturing, retail, schools, car washes) who want security plus operations and safety analytics, priced per location and sold through integrators since May 2026. Ambient.ai fits large enterprises with a security operations centre that need threat detection at scale; it says it delivers the most value from about 100 cameras.
• Does Eagle Eye Networks support AI video analytics such as motion detection, people counting and object recognition? Yes. Motion-based recording and alerts are part of the Cloud VMS, and in 2026 Eagle Eye lists people and vehicle detection, counting, line crossing, intrusion and loitering, license-plate recognition, gun detection, face match and natural-language video search. Several of these are add-on features, and they run on cameras connected through its Bridge or cloud recorder.
• How do you trial agentic AI without ripping out your existing VMS or NVR? Pilot an overlay. Ambient.ai and Spot AI both read existing IP cameras over ONVIF or RTSP through an on-site appliance, so one site with 20–100 cameras can run for 30–60 days while your current VMS stays the system of record; with Spot AI, run its recorder in parallel during the pilot. Measure false alarms per camera per day and time to respond before and after, and keep a human approving any action the agent takes.
• Eagle Eye Networks vs traditional CCTV: a traditional CCTV system records locally on a DVR or NVR you own and maintain; Eagle Eye moves recording, search and analytics to a per-camera cloud subscription and keeps your cameras.
• ZeroEyes vs Ambient.ai: ZeroEyes does one job, visible-gun detection confirmed by people in its own operations centre; Ambient.ai covers a broad library of threat signatures, of which weapons are one.
• Spot AI vs Avigilon: Avigilon under Motorola Solutions is an established VMS that sells its own cameras and adds analytics to a recording core; Spot AI is camera-agnostic software built around agents.

What "AI-native" actually means

Every VMS vendor says it has artificial intelligence now, so the phrase needs a precise meaning or it is just a sticker. The useful distinction is architectural, and it comes down to the order the system was built in.

The established platforms began life as recording software. Their first job — and the thing they are still superb at — was to take many camera streams, write them to disk reliably, and let an operator watch and search them. Analytics arrived later, as a module you add, a server you stand up, or a camera you buy. That is not a criticism; a recorder that never drops a frame under load is hard engineering, and we said so in the Milestone and Genetec profiles. But the artificial intelligence is a layer on top of a recording core.

An AI-native platform inverts that order. It was designed from the start so that understanding the video — detecting, classifying, reasoning about what is happening — is the primary job, and recording is in service of that. Think of the difference between a filing cabinet that someone later taught to read, and a reader that also keeps files. Both hold your documents; only one was built to understand them first. In practice "AI-native" almost always travels with "cloud-native," because the computing power to run modern vision models across hundreds of cameras is cheaper and more elastic to rent in a data center than to buy and rack at every site. The trade is that the video, or at least the analytics on it, now depends on a network link and a subscription.

Figure 1. Two build orders. The incumbent VMS started as a recorder and added artificial intelligence as a layer; the AI-native platform was built so understanding the video is the core job, with recording in service of it, cloud-managed and often with the processing at the edge. Neither is "better" in the abstract — they fit different buyers.

One honest caveat belongs here, before any vendor name. "AI-native" describes how a system is built, not how accurate it is. A platform designed around analytics still produces a precision-and-recall range — a realistic rate of correct detections against missed ones and false alarms — that depends on scene, lighting, angle, and tuning, exactly as every analytic does. Being analytics-first tends to mean the detection is better integrated and easier to use, not that it is perfect. We treat that reality in depth in tuning analytics: false alarms and accuracy, and we will hold every vendor in this profile to it.

Three platforms, three different animals

The biggest mistake a buyer makes with this category is treating the three as interchangeable "AI camera companies." They occupy meaningfully different positions, and the cleanest way to see it is to ask one question: what does this platform replace?

Eagle Eye Networks replaces your recorder and your VMS. It is a full cloud VMS — the system of record. You point cameras at it, it stores and manages the video, and operators live in its interface. If you adopt Eagle Eye, it is your VMS.

Spot AI layers onto the cameras you have and replaces the recorder with its own edge appliance. It started out selling its own AI cameras; in 2026 it describes itself as camera-agnostic software, with new cameras coming from certified hardware partners. Its pitch is that it is the intelligence layer — an on-site appliance reads the cameras you already own and turns them into what it calls Video AI Agents that not only watch but act.

Ambient.ai replaces nothing. It is a threat-detection layer that sits on top of the cameras and the VMS you already run, reads their feeds, and raises alerts. You keep your recorder; Ambient.ai is the brain you bolt onto it.

Hold that spectrum — overlay, intelligence layer, full platform — because it determines whether one of these is a rip-and-replace decision or an add-on, and it changes who else at the table needs to approve it.

Figure 2. What does it replace? Ambient.ai overlays the cameras and VMS you already own; Spot AI is camera-agnostic edge AI that layers on your cameras and replaces the recorder with its own appliance; Eagle Eye Networks is a full cloud VMS that becomes your system of record. The further right, the bigger the switch.

Eagle Eye Networks: the cloud VMS

Of the three, Eagle Eye Networks is the one that looks most like a traditional VMS — because it is one, rebuilt for the cloud. Understanding its history explains why it behaves the way it does.

Eagle Eye Networks was founded in 2012 by Dean Drako, who had previously founded and run Barracuda Networks, the company that shipped the IT-security industry's first spam-filter appliance and grew, from 2003 to 2012, to more than 200 million US dollars in annual sales and 150,000 customers. Drako brought an internet-infrastructure mindset to surveillance: build it in the cloud, secure it like an IT product, and expose it through a programming interface. The first product, the Eagle Eye Cloud Security Camera VMS, was announced in early 2014, and the platform has been a cloud VMS ever since rather than a desktop product later moved online.

The architecture answers the obvious objection to cloud surveillance — that you cannot push every camera's full-resolution video to a data center over an ordinary internet link. Eagle Eye offers two ways in. The first is the Bridge: an on-site appliance that sits between the cameras and the cloud, takes the heavy local video stream, isolates the cameras from the public internet for security, and sends up only what is needed — a model we explained as the edge-buffer pattern in cloud and hybrid storage for surveillance. The Bridge suits large single-site deployments with serious camera counts. The second is Camera Direct: newer cameras that connect straight to the cloud with no on-site appliance at all, for simpler or smaller sites. Either way, the system of record is the cloud, and recording can be kept in the cloud, on-premises, or both.

Two engineering choices make Eagle Eye truly open at the camera layer, which matters because cloud platforms are often assumed to be closed. First, it is camera-agnostic: it captures video and runs its analytics on cameras from across the market — its 2026 site cites more than 7,500 supported cameras — by speaking ONVIF, the common language that lets cameras and video software from different makers work together. Eagle Eye is an ONVIF member, and it can ingest and analyze video from effectively any ONVIF-conformant camera, which is why adopting it rarely means replacing your cameras. The mechanics of that standard are in ONVIF explained for engineers. Second, it exposes a developer-friendly web Application Programming Interface (API) — a published way for other software to pull video, events, and analytics in and out — which integrators have used to wire Eagle Eye into alarm monitoring, point-of-sale systems, and third-party dashboards.

The analytics follow the cloud-native promise: most run on the cameras you already have, with no special hardware. Eagle Eye's 2026 suite includes natural-language Smart Video Search, precision person and vehicle detection, counting, line crossing, intrusion and loitering detection, license-plate recognition (which it markets as working with ordinary cameras rather than dedicated plate cameras), face match, and gun detection. The gun detector, sold as Brivo Eagle Eye Gun Detection since October 2025, checks each detection three times — edge AI, a cloud model, then a human — and won the AI category of the 2026 Secure Campus Awards; the accuracy claims behind it are the vendor's own. Two of those — plate reading and face match — walk a deployment up to a legal gate, and we will return to that in the openness and privacy sections. Commercially, the platform is sold the cloud-native way: a usage-based subscription per camera rather than a one-time licence, bundling the recording, storage, and updates into a recurring fee, with several analytics sold as add-on features.

The most important current fact about Eagle Eye is corporate, not technical. In late December 2025, Eagle Eye Networks merged with Brivo, a cloud-based access-control company, to form what the two companies described as the industry's largest cloud-native physical-security business. The combined company operates under the Brivo name, with Dean Drako as chief executive; the Eagle Eye Cloud VMS becomes the video pillar of a broader "Brivo Security Suite" that spans access control, video, visitor management, and intrusion. For a buyer evaluating Eagle Eye today, that means two things: the cloud VMS itself continues, and the strategic direction is now a unified access-plus-video cloud platform rather than video alone. In January 2026 the leadership described the plan as one suite, one bill and one support contact, with the sales and R&D teams combined. As of October 2026 the video product is still sold as the Eagle Eye Cloud VMS on een.com, alongside Brivo Eagle Eye Video and Brivo Eagle Eye Complete bundles, so the brand you put on a shortlist may change name before your contract renews.

Figure 3. Eagle Eye Networks, end to end. ONVIF cameras reach the cloud either through an on-site Bridge appliance (large sites) or Camera Direct (simpler sites); the cloud VMS records, runs analytics on existing cameras, and serves operators and an open API. The system of record lives in the cloud.

Spot AI: camera-agnostic edge AI with agents that act

Spot AI starts from a different question than Eagle Eye. Eagle Eye asked, "what does a VMS look like if you build it in the cloud?" Spot AI asks, "what if the camera were a teammate that could act, not just a sensor that records?"

The company was founded in 2018 by three Stanford friends — Rish Gupta (now chief executive), Tanuj Thapliyal, and Sud Bhatija — who came out of the engineering teams at Cisco Meraki and Samsara, two companies known for making once-complex infrastructure simple and cloud-managed. That lineage shows in the product. Spot AI's hardware today is an on-site appliance, the Intelligent Video Recorder, which runs the vision models locally — in 2024 the company said it carries roughly three times the computing power of a traditional artificial-intelligence camera — so full-resolution video stays on site and only metadata goes to the cloud. It no longer positions itself as a camera maker: the appliance reads the IP cameras a business already owns over RTSP or ONVIF, regardless of vendor, and buyers who need new cameras are pointed to certified hardware partners. The result is an open intelligence layer that turns existing cameras into what Spot AI calls Video AI Agents.

The Video AI Agent is the differentiator, and it is worth defining plainly because the word "agent" is overused. A traditional analytic detects something and raises an alert for a human to handle. An agent, in Spot AI's sense, is given a goal and a set of allowed actions, and it both spots a situation and responds to it — alerting the right person, delivering an analytic, triggering lights or sounds, playing a voice message, or starting and stopping a machine if an incident or injury happens. Spot AI launched these agents in October 2024. In March 2026 it went a step further and repositioned them as “AI coworkers” that run whole workflows: watch a site, reason across camera feeds, company documents and access-control data, act, escalate and close the loop, with edge compute for real-time action and cloud reasoning for longer analysis. The named products are an AI Security Guard, an AI Operations Assistant, an AI Safety Manager (PPE, zone intrusion, vehicle-pedestrian conflicts) and Iris, a natural-language interface to the video; integrations go through open APIs, webhooks and a read-only Model Context Protocol (MCP) endpoint. As of October 2026 the company reports about 1,200 US customers, more than 2,200 active locations and more than 2,000 deployed agents, has raised more than 110 million US dollars (including a 31-million-dollar round led by Qualcomm Ventures in October 2024), and says it indexes more new video each day than is uploaded to YouTube. Two commercial facts matter to a buyer: since 1 May 2026 Spot AI sells only through channel partners, so your contract is with an integrator, and it prices per location rather than per camera.

The honest framing matters most here, because "an agent that acts on its own" is exactly the claim that needs accuracy-vs-performance discipline. Letting software take a physical action — sounding an alarm, stopping a production line — on the strength of a detection raises the cost of a false positive from "an operator glances at a clip" to "the line stopped for nothing," and the cost of a false negative from "a missed clip" to "the action that should have fired didn't." Spot AI's own customer figures are operational outcomes the vendor reports — a manufacturer cutting injuries, a school cutting incident-resolution time — and they are plausible and useful, but they are vendor-reported results, not independent benchmarks, and the underlying detection is still a precision-and-recall range. The right way to deploy an acting agent is with a human in the loop for consequential actions and a tuned operating point, a discipline that is not optional, and for high-risk AI systems under the EU AI Act's human-oversight rules (Article 14) a legal expectation, which we cover in tuning analytics: false alarms and accuracy. The agent model is a real advance and a useful one; it also concentrates the consequences of being wrong, so it deserves the most careful piloting of anything in this profile.

Ambient.ai: the threat-detection layer that replaces nothing

Ambient.ai is the purest expression of "analytics-first," because analytics is all it is. It does not want to be your VMS or sell you cameras; it wants to be the intelligence that sits on top of whatever you already run.

The company was founded in 2017 by Shikhar Shrestha (chief executive) and Vikesh Khanna (chief technology officer) and came out of stealth in early 2022 with a platform it calls computer-vision intelligence. The idea is to retrofit into a building's existing cameras and alarm sources, continuously read those feeds, and recognize not just objects but situations — a person tailgating through a door, a weapon being drawn, someone loitering at a perimeter, a vehicle following another too closely. Ambient.ai describes these as "threat signatures": patterns of behavior and context that, when matched, raise a real-time alert so a security team can respond before an incident escalates rather than reviewing it afterward. Because it layers onto existing infrastructure, adopting Ambient.ai is an add-on decision, not a rip-and-replace, which is a meaningfully smaller commitment than swapping a VMS. In practice it runs as a hybrid: edge GPU appliances on site do the perception and keep the raw video in your environment, and a cloud console serves the security operations centre. It integrates with Genetec and Milestone VMSs, with 10+ access-control systems, and with ONVIF cameras, and the company says it delivers the most value from about 100 cameras.

Two things make Ambient.ai distinctive. The first is a deliberate privacy stance: the company is publicly built around not using facial recognition, and says it will never add it, positioning context-and-behavior analysis as a way to detect threats without identifying individuals by their biometrics. That is a deliberate design choice with real consequences, and it sidesteps the heaviest part of the legal gate we describe below — though it does not remove every privacy obligation, because detecting and recording identifiable people is still processing personal data. The second is the reasoning model. In November 2025 Ambient.ai made Ambient Pulsar generally available: a reasoning vision-language model — an artificial-intelligence model that can describe and reason about a scene in something close to natural language — that runs on its on-site edge appliance, was trained on more than a million hours of enterprise video, and is built for open-set detection of behaviours it was not explicitly trained on. On top of it came natural-language Activity Notifications and Semantic Search over the archive and, in August 2026, Agentic Video Walls, AI-written case narratives and twice as many camera streams per edge appliance. Its comparisons against GPT-5 and Gemini 2.5 Pro are the company's own, not independent benchmarks. Commercially it is growing fast: new annual recurring revenue doubled in its fiscal 2026 (ending January 2026), and on 20 October 2026 Allegion Ventures led its Series C, taking total funding past 100 million US dollars. That is the same agentic frontier Spot AI is pushing toward from the edge-appliance side, and the clearest signal of where this whole category is going.

The accuracy discipline applies in full. A system whose entire value is catching threats lives and dies on the balance between catching real ones (recall) and not crying wolf (precision). A threat-detection layer that floods a security team with false alarms gets ignored, and one tuned too conservatively misses the event it was bought to catch. Ambient.ai's pitch is precisely that its context modeling reduces false alarms relative to naive motion alerts — a real and valuable claim — but "fewer false alarms" is not "no false alarms," and the deployment still needs a tuned operating point and human verification on consequential alerts. We treat the rule-free, learn-what-is-normal style of detection it exemplifies in anomaly detection in surveillance video.

Openness and the standards boundary: the lock-in moved

Here is the counterintuitive thing about cloud, AI-native platforms, and the single most useful idea in this article. They are usually more open at the camera than the incumbents — and more locked-in at the layer that matters.

All three lean on the open camera standard to get video in. Eagle Eye is camera-agnostic over ONVIF; Spot AI advertises that it works with existing cameras regardless of vendor; Ambient.ai retrofits into whatever cameras you already have. Remember, ONVIF only guarantees a baseline — the common features a conformant camera and the platform both support, enough to pull a stream and basic events — and we keep "ONVIF-conformant" and "fully featured" as separate ideas throughout this section. But for the job these platforms need from a camera, the standard is enough, which is why none of them requires you to rip out your cameras. That is a real advantage over an appliance vendor whose best features assume its own hardware.

So where is the lock-in? It moved up the stack. With an AI-native cloud platform, the thing you depend on is not the camera and not even the recorder — it is the cloud service, the subscription, and the accumulated data and analytics living inside the vendor's platform. Your video history, your tuned agents, your threat signatures, your search index: with Eagle Eye these live in the vendor's cloud; with Spot AI and Ambient.ai the raw video stays on your site, but the models, agents and index are still proprietary, reachable through the vendor's API on the vendor's commercial terms. Switching cameras is easy; switching platforms means migrating all of that. The open API softens this — it is markedly easier to get your data and events out of these platforms than out of a closed appliance — but the gravity is the cloud, not the hardware. A senior engineer should read this as: the integration surface is excellent and standards-based at ingest, and proprietary at the point of value. Price the exit, not just the entry.

Figure 4. The lock-in moved up. AI-native platforms ingest video over the open ONVIF standard and expose open web APIs, so the camera layer is truly portable; the proprietary cloud, the analytics, the tuned agents, and the data history are where the switching cost lives. Open at the camera, locked at the cloud.

The privacy gate sits on the same boundary and deserves a plain statement. Cloud analytics that detect people, plates, or faces process personal data, and the moment a platform is configured for biometric face matching it crosses into special-category data under the European Union's General Data Protection Regulation (GDPR Art. 9) and, in Illinois, the consent-and-private-right-of-action regime of the Biometric Information Privacy Act (BIPA, 740 ILCS 14). Eagle Eye's plate reading is personal data governed by retention and access rules; Ambient.ai's no-facial-recognition stance is partly a way to stay clear of the heaviest gate; any platform's face-matching feature is a legal decision before it is a technical one. We treat the law itself in GDPR for video surveillance, the biometric gate in face recognition in surveillance, and plates in license-plate recognition. The point here is that an AI-native platform makes these capabilities a checkbox, so know where the gate is before you tick it.

The cost shape: renting capability instead of owning it

AI-native platforms almost all share one commercial shape, and seeing it clearly is worth more than any feature list. They are an operating expense — a recurring subscription, per camera (Eagle Eye) or per location (Spot AI) — where the on-premises incumbents are a capital expense.

The on-premises model, which we walked through for Avigilon Unity and the others, is a one-time purchase: you buy the camera licences, you buy and rack the servers and storage, and you own it. The AI-native cloud model flips this. You pay a recurring fee per camera that bundles the recording, the storage, the analytics, and the updates, and you own no infrastructure. Walk the arithmetic out loud, because the shape is the point, not the exact figures. Take a 100-camera site and an illustrative per-camera cloud subscription:

annual_subscription = price_per_camera_per_month × cameras × 12
annual_subscription = $30 × 100 × 12
annual_subscription = $36,000 per year   (recording, storage, analytics, updates included)

Over five years that is roughly 180,000 US dollars, with no servers to buy and nothing to maintain — but it never stops, and it scales linearly with cameras and retention. The on-premises alternative front-loads a large capital cost and then runs cheaper year to year, which is why the honest comparison is never "which is cheaper this month" but "which shape fits how we budget and how long we keep the system," compared on the same multi-year footing. We build that whole-system comparison in the surveillance cost model.

Figure 5. The two cost shapes. The on-premises purchase is a big up-front capital cost that then climbs gently; the AI-native subscription starts low and climbs steadily, typically overtaking the one-time purchase around year three. Neither is "cheaper" in the abstract — they fit different budgets and time horizons.

Two cloud-specific costs hide inside the subscription and surprise teams that do not plan for them. The first is bandwidth: a cloud platform has to get video off-site, and continuous upload of many cameras can exceed an ordinary business internet link, which is exactly why Eagle Eye's Bridge and Spot AI's appliance buffer locally and send up less — the upload wall we quantify in cloud and hybrid storage for surveillance. The second is the cost and latency of cloud analytics at scale, where sending more video up for processing meters up quickly; we model that in cloud video analytics cost and the broader edge-vs-cloud deployment decision. None of this makes cloud wrong — it makes it a different cost shape to plan, not a cheaper one to assume.

A common mistake to avoid

The costliest errors with AI-native platforms come from believing the label instead of testing the system. Four recur. First, assuming "AI-native" means "more accurate": it means analytics-first by design, not perfect detection — every one of these platforms produces a precision-and-recall range that you must verify on your own scenes and lighting, never a guaranteed number. Second, letting an agent act without a human in the loop on consequential actions: an autonomous response amplifies both correct and incorrect detections, so a stopped line or an ignored alarm is the predictable cost of skipping the oversight step — which is also a legal expectation, not just good practice. Third, forgetting the cloud is a dependency: if the platform's value is in the cloud and the site's internet link or the service goes down, plan for what keeps recording locally and what degrades gracefully. Fourth, treating biometric features as ordinary settings: switching on face matching is a legal decision under GDPR Art. 9 and BIPA before it is a configuration, and "the platform supports it" is not the same as "you may deploy it here." None of these is a flaw in the products; each is a planning gap an honest evaluation closes before the pilot, not after.

The three versus the incumbents, without spin

The most useful thing a profile can do is place these platforms side by side with the established VMSs from earlier in this block, on the dimensions a buyer actually weighs. All are serious products; the choice is about deployment model, openness, and where the intelligence and the lock-in live.

Table 1. AI-native platforms versus the incumbents, on the dimensions a buyer weighs. The fork is not "AI or no AI" — every row has AI now. It is what the platform replaces, where it runs, and whether the intelligence was the starting point or a later layer.

The plain reading: choose Eagle Eye Networks when you want a cloud VMS as your system of record across many sites, keep your existing cameras, and value an open API and usage-based pricing — and you are comfortable that the company is now inside Brivo's access-plus-video strategy. Choose Spot AI when the job reaches past security into operations and safety, the idea of agents that take actions, not just raise alerts, solves a real problem you can pilot carefully, and you are happy to buy through an integrator. Choose Ambient.ai when you want to keep the VMS you already run, add proactive threat detection on top, and value a vendor that deliberately avoids facial recognition. Choose an incumbent when a proven on-premises recorder, maximal third-party integration, or full local control outweighs the analytics-first, cloud-first pitch. Whether to adopt any platform, assemble on open components, or build is the subject of custom vs off-the-shelf VMS, and how to weigh any comparison like the one above is in reading a VMS comparison.

Where AI-native fits — and where it doesn't

An honest profile names both sides. The analytics-first, cloud-first platforms fit when proactive intelligence and low operational friction are the job: a multi-site retailer, a fast-growing operations team, or a security group that wants threats surfaced in real time without standing up and maintaining servers at every location. They fit when you want to keep your existing cameras, value an open API for integration, and prefer a predictable subscription, per camera or per location, over a large capital purchase. And they fit especially well when the use case reaches beyond security into operations and safety, where Spot AI's agents and the wider agentic direction have the most to offer.

They are the wrong tool when the constraints point the other way. A site with poor or expensive connectivity, strict data-residency rules that forbid sending video off-premises, or a hard requirement to keep recording through any internet outage will be better served by an on-premises or hybrid design — the trade-offs are in on-prem, cloud, and hybrid VMS. An organization that needs maximal third-party device integration or deep on-prem control may find a mature incumbent fits better. And any buyer tempted by "AI-native" as a guarantee of accuracy should re-read the first section: it is a description of architecture, and the detection still has to be tuned and verified on your scenes. Analytics-first, cloud-first surveillance is, by the same token, the most network-dependent and the most subscription-bound — which is exactly the trade an honest evaluation weighs against the proven, on-prem alternatives profiled earlier in this block.

For the commercial overview of the market this profile sits inside, see Fora Soft's video surveillance management systems playbook and the rundown of modern VMS software features.

Where Fora Soft fits in

Fora Soft has built real-time video, streaming, and computer-vision software since 2005, across 250+ shipped projects, and a large share of our surveillance work lives exactly where an AI-native platform meets a customer's own systems — pulling video and analytics events out of a cloud VMS through its API into bespoke dashboards and business workflows, building the specific detection a packaged platform does not offer, or wiring an analytics overlay into an existing recorder. The discipline we bring is the one this section preaches: design for how the system behaves at full camera load and on a bad-network day first — realistic detection precision and recall under real lighting, latency you have measured, recording that degrades gracefully when the internet link drops — then the feature list. When a team sits between "subscribe to an AI-native platform" and "build the intelligence ourselves," we help weigh extending an open cloud platform through its API against assembling on components, with the honesty that a reliable cloud VMS and a tuned analytics pipeline are expensive to reinvent and the better vendors here solved a lot of it already. When the answer is to build, our video surveillance software development team owns the VMS layer, and our computer vision for surveillance engineers ship the detection models that run on it at the edge. If what you need is an agent that reasons over live video and triggers actions — the Spot AI and Ambient Pulsar pattern, but on your own cameras, rules and data — our AI video agent development team builds it on YOLO-class detectors and NVIDIA Jetson at the edge, with a vision-language model for the reasoning step.

The handbook

Fora Soft Video Engineering Handbook

We wrote a seven-volume series that takes the whole video stack apart across 9,000+ pages — encoding, delivery, audio and AI. Volume 1 covers the encoding that sets what a camera stream costs to send and store, and volumes 6 and 7 cover the vision models, the real-time serving cost and the compliance work sitting behind every detection claim in this profile. Work through them and you can design and build a platform like the ones profiled here, not only choose between them. Written by Nikolay Sapunov, CEO at Fora Soft.

Browse the handbook on Amazon →7 volumes · Kindle & paperback

See our case studies

References

1. ONVIF — "ONVIF Profiles" and "Member Profile: Eagle Eye Networks" (an ONVIF profile is a fixed set of features a conformant device and client must support; conformance to a profile is what ensures interoperability between ONVIF products; Eagle Eye Networks, as an ONVIF member, can capture video and AI analytics from virtually any ONVIF-conformant camera — the open camera-ingest baseline every cloud VMS in this profile relies on). Primary standard (tier 1). https://www.onvif.org/profiles/
2. European Union — "General Data Protection Regulation (Regulation (EU) 2016/679)" (Art. 9 treats biometric data used to uniquely identify a person as special-category data; Art. 35 requires a Data Protection Impact Assessment for high-risk processing such as large-scale systematic monitoring — the legal frame any face-matching or plate-reading feature in a cloud analytics platform must answer to). Primary law (tier 1). https://eur-lex.europa.eu/eli/reg/2016/679/oj
3. Illinois General Assembly — "Biometric Information Privacy Act (740 ILCS 14)" (requires informed written consent before capturing a face template or other biometric identifier, sets a retention-and-destruction duty, and provides a private right of action with statutory damages — the heaviest US biometric gate, the one Ambient.ai's no-facial-recognition stance is partly designed to stay clear of; statute text at ilga.gov, linked here via an encyclopedic summary because the legislature's site blocks automated checks). Primary law (tier 1), encyclopedic link (tier 4). https://en.wikipedia.org/wiki/Biometric_Information_Privacy_Act
4. United States Congress — "John S. McCain National Defense Authorization Act for Fiscal Year 2019, Section 889 (Public Law 115-232)" (bars federal agencies, their contractors, and grant recipients from procuring video-surveillance equipment from named manufacturers on national-security grounds; a camera-procurement gate any cloud VMS deployment in a government or federally funded setting must clear, even when the VMS itself is camera-agnostic). Primary law (tier 1). https://www.govinfo.gov/app/details/PLAW-115publ232
5. Business Wire / Brivo — "Brivo and Eagle Eye Networks Merge to Form the Industry's Largest AI Cloud-native Physical Security Company" (29 December 2025: Eagle Eye Networks, founder Dean Drako's cloud VMS, merges with cloud access-control company Brivo; the combined company operates under the Brivo name with Drako as CEO and Steve Van Till as President; the Brivo Security Suite spans AI, access control, video intelligence, visitor management, and intrusion — the single most important current fact about Eagle Eye). First-party press (tier 3). https://www.een.com/blog/brivo-and-eagle-eye-networks-merge-to-form-the-industrys-largest-ai-cloud-native-physical-security-company/
6. Brivo — "Brivo Eagle Eye Gun Detection Wins 2026 Secure Campus Award" (14 July 2026: gun detection introduced October 2025, triple-layer verification — edge AI, cloud model, human; works with existing cameras; product naming after the merger: Eagle Eye Cloud VMS, Brivo Eagle Eye Video, Brivo Eagle Eye Complete, Brivo Security Suite). First-party press (tier 3). https://www.brivo.com/resources/blog/brivo-eagle-eye-gun-detection-wins-2026-secure-campus-award/
7. Security Systems News — "'We are set to lead again': Drako on Brivo, Eagle Eye merger" (22 January 2026: one suite, one bill and one support contact; sales and R&D teams to be combined). Trade press (tier 5). https://digital.securitysystemsnews.com/articles/-we-are-set-to-lead-again-drako-on-brivo-eagle-eye-merger
8. Eagle Eye Networks — "Eagle Eye Cloud VMS" product and API documentation (a cloud-native VMS with cloud and on-premises recording, the on-site Bridge appliance and direct-to-cloud Camera Direct, support for 7,500+ cameras, usage-based per-camera subscription pricing, an open RESTful Video API, and analytics — Smart Video Search, LPR, gun detection, person/vehicle detection, occupancy, perimeter intrusion — that run on existing cameras). First-party engineering (tier 3). https://www.een.com/
9. Eagle Eye Networks — "Barracuda Networks Founder & Former CEO Dean Drako Launches Eagle Eye Networks" and company history (Eagle Eye founded 2012 by Dean Drako, who built Barracuda Networks from 2003 to 2012 to over US$200M in annual sales and 150,000 customers with the industry's first spam-filter appliance; the Eagle Eye Cloud Security Camera VMS was announced in early 2014 — the cloud, IT-security lineage of the platform). First-party / encyclopedic (tier 4). https://en.wikipedia.org/wiki/Eagle_Eye_Networks
10. Spot AI — "Spot AI Introduces First Video AI Agents for the Physical World as it Nears $100 Million in Funding to Date" (29 October 2024: Spot AI, founded 2018 by Rish Gupta, Tanuj Thapliyal, and Sud Bhatija out of Cisco Meraki and Samsara engineering teams, launches Video AI Agents that act on detections — alerting, triggering lights/sounds, starting/stopping machines; an edge appliance with ~3× the compute of a traditional AI camera; an open system that works with existing cameras). First-party press (tier 3). https://www.spot.ai/blog/spot-ai-introduces-first-video-ai-agents-for-the-physical-world-as-it-nears-100-million-in-funding-to-date
11. Spot AI — "Spot AI Goes 100% Channel, Launches Industry-First $2M Equity Program as Video AI Evolves from Task Completion to Full AI Coworker" (25 March 2026, ISC West: all deals on partner paper from 1 May 2026; agents repositioned as AI coworkers that run whole workflows with edge compute plus cloud reasoning; 1,000+ customers; over $110M raised). First-party press (tier 3). https://www.spot.ai/blog/spot-ai-goes-100-channel-launches-industry-first-2m-equity-program-as-video-ai-evolves-from-task-completion-to-full-ai-coworker
12. Spot AI — "Official information about Spot AI" (updated October 2026: camera-agnostic software, does not manufacture cameras, certified hardware partners; Intelligent Video Recorder edge appliance, full-resolution video on site and metadata to the cloud; RTSP/ONVIF ingest; about 1,200 US customers, 2,200+ locations, 2,000+ agents; AI Security Guard, AI Operations Assistant, AI Safety Manager, Iris; open APIs, webhooks, read-only MCP endpoint; per-location subscription; SOC 2 Type II, NDAA). First-party (tier 3). https://www.spot.ai/ai-instructions
13. TechCrunch — "Ambient.ai aims to provide AI-powered building security, minus bias and privacy pitfalls" (19 January 2022: Ambient.ai, founded 2017 by Shikhar Shrestha and Vikesh Khanna, emerges from stealth with a 'computer vision intelligence' platform that retrofits into existing cameras and uses 'threat signatures' to detect threats contextually, deliberately without facial recognition to avoid bias and privacy pitfalls — the no-FR design stance and the overlay model). Institutional reporting (tier 5). https://techcrunch.com/2022/01/19/ambient-ai-security-without-facial-recognition/
14. Ambient.ai — "Ambient.ai Raises $20M Strategic Growth Investment from Allegion Ventures" (30 October 2023: a $20M strategic growth investment, described as the fund's largest to date). First-party press (tier 3). https://www.ambient.ai/press/ambient-ai-strategic-growth-investment-allegion-ventures
15. Ambient.ai / PR Newswire — "Ambient.ai Unveils Pulsar, the first Reasoning Vision-Language Model for Agentic Physical Security" (19 November 2025: general availability; runs on the on-site edge appliance; trained on 1M+ hours of enterprise video; open-set detection; Agentic Video Walls, Activity Notifications, Semantic Search; benchmark claims are vendor-reported). First-party press (tier 3). https://www.prnewswire.com/news-releases/ambientai-unveils-pulsar-the-first-reasoning-vision-language-model-for-agentic-physical-security-302619452.html
16. Ambient.ai / PR Newswire — "Ambient.ai Doubles New Annual Recurring Revenue as Agentic Physical Security Reaches Inflection Point" (18 February 2026: new ARR in FY26, ending 31 January 2026, doubled versus FY25; net revenue retention 140%+). First-party press (tier 3). https://www.prnewswire.com/news-releases/ambientai-doubles-new-annual-recurring-revenue-as-agentic-physical-security-reaches-inflection-point-302691381.html
17. Ambient.ai — "Ambient.ai Introduces New Agentic Physical Security Capabilities for Monitoring and Investigations" (26 August 2026: Agentic Video Walls, Case Management with AI-written narratives, grouped Semantic Search results, up to 2× camera streams per Ambient Edge Appliance). First-party press (tier 3). https://finance.yahoo.com/technology/ai/articles/ambient-ai-introduces-agentic-physical-130000192.html
18. Allegion Ventures / Business Wire — "Allegion Ventures Makes Follow-On Investment in Ambient.ai, a Leader in Agentic Physical Security" (20 October 2026: Allegion Ventures leads Ambient.ai's Series C; total funding over $100M; 90% ARR growth in FY2026; cash-flow positive in H1 2026). First-party press (tier 3). https://www.financialcontent.com/article/bizwire-2026-10-20-allegion-ventures-makes-follow-on-investment-in-ambientai-a-leader-in-agentic-physical-security
19. Ambient.ai — "AI info" company facts page (updated April 2026: no facial recognition and a commitment never to add it; Genetec and Milestone integrations, 10+ access-control systems, ONVIF cameras; hybrid edge plus cloud SOC; most value from about 100 cameras; SOC 2 Type II). First-party (tier 3). https://www.ambient.ai/ai-info
20. IEC — "IEC 62676 series: Video surveillance systems for use in security applications" (specifies minimum requirements across the system lifecycle; EN IEC 62676-4:2025 covers application guidelines including information security and data privacy — the system-level floor any VMS, cloud or on-prem, should meet). Primary standard (tier 1). https://webstore.iec.ch/en/publication/34391