What Vidan AI's own documentation covers, and what it leaves to a conversation, written for IT and operations teams running due diligence on a configurable video analytics platform. Every constraint carries its documentation, its operational impact, and its workaround.
Every constraint below comes from Vidan AI's own current documentation, checked on August 17, 2026. More of this page than usual reports what the documentation does not cover, and each of those is named as an absence rather than turned into a claim about the product, because an absence is a question to ask. Spot AI sells a competing platform, so nothing here rests on an anonymous source or an aggregated user rating, and nothing here should be read as evidence that an undocumented capability is missing. Spot AI has published its own page about Vidan AI, and this check found nothing in it that Vidan's current documentation contradicts, so this banner carries no correction.
The premise and one worked example are documented well enough to judge. The mechanics underneath them are not on the pages at all, which makes the first call a specification exercise rather than a demo.
A constraint list is only useful next to an honest account of the product.
Vidan publishes that it gives you complete control over what matters most by letting you define specific objects and actions to monitor, with detection rules flexible per location, camera angle and operational need. The examples are the kind an operations lead actually raises: unauthorized vehicles, critical tools, personal protective equipment compliance, running in restricted areas, a secure door being opened and prolonged inactivity in a high-risk zone.
Trans-Global Solutions runs what Vidan calls a custom-built system for freight: cameras track freight trains through the transport cycle, and the detection focuses on the caps used to load and seal liquid cargo, which is exactly where a spill starts. An unsecured cap raises an alert so the team can correct it, and operators can flag anything the system missed so the model improves. Vidan states more than 10,000 caps monitored.
A dedicated machine learning engineering practice covers model design through deployment, with continuous performance tracking, early issue detection and ongoing retraining named as the service, and Vidan is candid about the failure modes: models that work in testing and break on real-world data, messy inputs, scaling problems and drift after deployment. A traffic control product covering crashes, wrong-way vehicles, congested intersections, violations and signal timing sits alongside it.
Each one is a consequence of how the platform is presented rather than a defect. What matters is whether it collides with your starting point.
No protocol, resolution floor, frame-rate range or analog route appears anywhere. The closest published statements are that Vidan turns ordinary cameras into smart security tools and works without disrupting your current security infrastructure. Those point the right way for a brownfield site, and neither is a specification an engineer can check a fleet against.
On an estate with a decade of mixed hardware, camera compatibility is the question that decides feasibility before anything else is worth discussing. A yard with analog runs on a hybrid recorder and IP cameras from three manufacturers cannot tell from public pages whether half the fleet participates, so the shortlist stalls at the first technical review.
Ask directly for the protocol, the minimum resolution and frame rate, and whether analog channels are reachable through an encoder, and get the answer in writing rather than in a demo. Then run a proof of concept on your oldest cameras rather than your newest, because those are the ones that decide whether the platform covers the estate or a slice of it.
Nothing states whether processing happens on site or in the cloud, what has to be installed, where full-resolution video ends up, how long it is kept or what happens to it at the end of a contract. Encrypted archives and secure processing controls are the published words. No SOC 2, ISO 27001 or NDAA statement appears beside them.
Encryption is an intention rather than a design a reviewer can assess, so an enterprise security team has nothing to read before it starts asking. On a site with an evidence obligation or a regulator, retention and export are contractual questions, and discovering that they are unwritten late in a cycle is what turns a preferred vendor into a delayed one.
Request four documents in one email at the start: an architecture and data-flow diagram, the retention schedule and export path, the end-of-contract deletion terms, and whatever certification pack exists. Ask early, and treat the answers as the evaluation rather than reading anything into their absence from a marketing page.
The premise is that you define specific objects and actions to monitor, and Vidan publishes a machine learning engineering practice covering model design, deployment, monitoring and retraining. What is not stated anywhere is who builds a given customer's detection, how long it takes, or whether that work sits inside the platform or alongside it as an engagement.
On a platform whose whole value is bespoke detection, that is the question a project plan depends on. Three detections at three sites could be a configuration exercise or three modeling projects with their own data collection and tuning, and the difference decides whether this is a quarter of work or a year of it. Retraining after drift is the same question again, every year.
Get the build path in writing: who does the work, how long a first detection takes, what a change costs in elapsed time, and who owns retraining after drift. Use the Trans-Global Solutions deployment as the benchmark to ask about, and agree a written acceptance test per detection before any commitment rather than after the first model lands.
The published response is a real-time event alert to the team plus a centralized monitoring dashboard across sites, with integration into existing alarms, access control and operational tools stated in general terms. No speaker, strobe, talk down or automated on-site response appears on any page. The documentation covers what the team is told, not what happens where the event is.
For a rail yard, a dealership lot or a construction site, the value is usually somebody being warned off rather than an incident being recorded well. A platform that raises a clean alert into a dashboard nobody is watching at 03:00 is producing a good record of something that still happened, and the deterrence budget then sits with a separate contract.
Name the response requirement as its own line in the evaluation, and ask which alarm or public address system Vidan can trigger and how. Where sites already have speakers or strobes, ask each option on the shortlist which of them drives that hardware directly, and price the answer alongside the detection work rather than after it.
The Trans-Global Solutions page states 99 percent detection accuracy and 200 percent ROI in months, and figures across the site include a 30 percent drop in intersection accidents, incident response 90 percent faster and safety incidents down by up to 65 percent. None carries a customer, a period, a baseline or a sample size. Trans-Global Solutions is the only customer named anywhere, and a line reading trusted by 100,000+ organizations globally appears with no organizations named beside it.
None of that makes the numbers wrong, and it does make them untestable, so they cannot be used as an acceptance threshold or compared against another vendor's figure. A buyer taking this to a steering committee has one named reference and a set of percentages, which is thin evidence for a multi-site commitment even when the product is a good fit.
Ask for the baseline, the observation window and the sample behind the 99 percent accuracy figure, and agree in writing what accuracy means before a pilot starts. Request three reference calls with customers in your own segment, and treat every percentage on the site as a claim to test on your cameras rather than as a forecast.
The same five constraints in one view, sized to paste into an evaluation document.
Swipe the table sideways to see every column.
Vidan AI data comes from Vidan AI's own product, industry, case study and machine learning engineering pages, checked on August 17, 2026. Gaps are marked as not publicly specified.
If the thing that has to be watched is specific to your operation, a cap on a rail tanker, a tool that must not leave a bay, a door that must not open during a shift, then a platform whose premise is that you define it is worth a conversation on that basis alone, and the Trans-Global Solutions page shows the model working on exactly that kind of problem. Roadway and smart-city work is the other case, because traffic control is a published product here rather than an afterthought. Both assume an evaluation that can run on a pilot and a set of conversations rather than on documents.
Spot AI answers the same bespoke requirement from a documented starting point, which is the main difference between the two. 15+ pre-trained Video AI Agents ship with the platform covering vehicle break-in, fire, cash register theft, after-hours intrusion, personal protective equipment, forklift near-miss, falls and crowding in hazard zones, so most of a site's list is already built, and Iris creates a custom detection in natural conversation in about eight minutes rather than through a modeling engagement with an unstated timeline.
The parts an evaluation has to read are published. Any ONVIF or RTSP IP camera works at full functionality and legacy analog cameras come in through the Intelligent Video Recorder, so feasibility on a mixed estate is a protocol question rather than a call. Full-resolution video stays on the IVR in the building and only event metadata crosses the network, and an event is answered on site with talk down, strobes and horns through standard speakers instead of stopping at a dashboard.
The useful question is not which platform has fewer constraints, but how much of your evaluation can be answered before the first call.
None of that makes Spot AI the right answer for every requirement. A genuinely unusual object, a cap on a rail tanker among them, may still need a conversation about whether Iris can be pointed at it, so a platform built entirely around bespoke modeling can reach places a shipped agent set does not. Spot AI publishes no traffic control or roadway product at all, so intersections, wrong-way vehicles and signal timing are outside its scope entirely, where Vidan sells them as a product. Spot AI does not publish a detection accuracy figure either, so on that specific number there is less to hold it to. A constraint list earns its keep by lining each platform's shape up against your starting point.
A live pilot on your cameras answers in a week what a spec sheet cannot.
Customer-reported outcomes from named Spot AI customers.
Wayne Transports covers terminals for an 800-vehicle fleet, documenting a $5,000 fuel-island collision on camera and running plate recognition without specialized hardware.
Unique Industries covers more than a million square feet with a three-person safety team, catching near misses and falls on the cameras already installed.
Eureka College cut footage review from hours to about five minutes across a 70-acre campus, with Brivo access control integrated alongside it.
"Our machines run 24/7, whether the lights are on or not. Spot AI lets us see what went wrong inside a machine even during those lights-out hours."
Five, and most of them are absences in the published documentation rather than published capabilities. Camera compatibility is not specified: no protocol, resolution floor or analog route appears. The architecture is not specified either, and no certification statement appears. Who builds a custom detection, and how long it takes, is not stated. The documented output stops at an alert and a dashboard, with no on-site response. And the published outcome figures carry no baseline, window or sample, with one customer named anywhere on the site.
Not publicly specified, and that is the finding rather than a criticism. The closest published statements are that Vidan turns ordinary cameras into smart security tools and works without disrupting your current security infrastructure, which point the right way for a site with hardware already installed. No protocol, resolution floor, frame-rate range or analog route appears anywhere, so the protocol, the minimum resolution and the analog question all have to be asked directly before feasibility can be judged.
The premise is that you define the specific objects and actions to monitor, with rules flexible per location and camera angle, and Vidan publishes a machine learning engineering practice covering model design, deployment, monitoring and retraining. What is not published is who does that work for a given customer, how long a first detection takes, or whether it is included or engaged separately, so those three answers belong in writing before a project plan is built around them.
They are published without the context needed to test them. The 99 percent detection accuracy and 200 percent ROI in months figures on the Trans-Global Solutions page carry no baseline, observation window or sample size, and the same is true of the percentage outcomes across the rest of the site. Treat them as claims to verify in a pilot on your own cameras and lighting, agree in writing what accuracy means before that pilot starts, and ask for references in your own segment.
It depends on how unusual the requirement really is. If it is genuinely one of a kind, a platform built around customer-defined models is the right shape and Vidan publishes a real worked example of one. If most of the list is standard and only one or two items are unusual, a platform such as Spot AI covers more before any custom work starts: 15+ pre-trained Video AI Agents on any ONVIF or RTSP camera plus legacy analog through the Intelligent Video Recorder, with Iris building the remainder in about eight minutes.