What AI Bot Eye's own site says its fire early-warning layer does not do, written for plant, hotel and facility teams whose cameras and recorders are already in place. Every constraint below carries the documentation behind it, the operational impact, and the workaround.
Every constraint below comes from AI Bot Eye's own current product and solution pages at aiboteye.com, checked on August 17, 2026. Where AI Bot Eye documents nothing, this page says so rather than guessing, and one of those silences is the sharpest finding here. Spot AI sells a competing platform, so nothing rests on a software-directory listing or an anonymous rating, and one claim that circulates about small vision vendors is corrected: AI Bot Eye publishes the boundary of its own product in plain language, which is more than most of this category does.
AI Bot Eye publishes both halves of its own alerting story, and they fail differently. The offline path carries the siren and the text message; the path carrying the image needs the network.
A constraint list is only useful next to an honest account of the product.
The site states that AI Bot Eye is an extra early-warning layer alongside certified fire systems and not a replacement for certified systems. A facilities manager needs that sentence before any feature list, and almost nothing at this end of the market volunteers it. A vendor that writes down where its job ends is a vendor whose other claims can be read at face value.
Feeds from Hikvision, CP Plus, Dahua, Uniview, Prama, Honeywell, Bosch and Hanwha are named, alongside most other DVR, NVR and IP systems. Detection is stated to be adapted to each camera view for angle, distance, lighting, activity and known patterns rather than dropped on every feed identically, and the site points to more than 50 live site demonstrations plus named deployments including The Leela Palace New Delhi.
Processing can stay on site, including on isolated recorder networks, without cloud video processing or continuous internet, and the site states that SMS and the hooter can work without internet. Alerts reach a control-room dashboard carrying camera location and snapshots, WhatsApp, SMS, phone calls and a local siren or hooter wired through GPIO, RS485, Modbus or dry contact relays.
Each one is a consequence of how the product is scoped rather than a defect. What matters is whether it collides with your starting point.
AI Bot Eye states that it is an additional visual early-warning layer alongside your certified fire systems, and that it does not replace certified life-safety systems. The published job is detecting visible fire early through camera views and alerting the team faster, with detection typically in under 5 seconds and a demonstration timed at 03.18 seconds.
The layer is additive, which means it is a second alerting path on top of a compliance obligation you already carry and already pay for. A plant hoping to retire a panel, a loop or a monitoring contract on the strength of this purchase is reading the site wrong, and the finance case has to be built on faster response rather than on removing a line item.
Budget it as an addition and keep the certified system in the same architecture diagram, so nobody downstream mistakes one for the other. AI Bot Eye's own answer is proof rather than argument: it publishes more than 50 live site demonstrations and invites a test on your own cameras, so price the early-warning claim against the minutes it actually saves.
The published claim is early detection of visible fire and smoke from the camera views already covering critical areas, with models adapted per view. The site is specific that the trigger is fire becoming visible in an existing camera view, and it describes controlled fires detected indoors, outdoors and in underground cellars.
That is the honest description of the technique and also its limit. A fire inside a machine enclosure, above a suspended ceiling, inside a duct or behind full racking is not a visible-flame problem until it is a large one, so camera placement stops being an installation detail and becomes part of the specification. On a warehouse where the aisles are covered and the mezzanine plant room is not, the coverage map is the product.
Walk the risk register against the camera plan before the pilot and mark every high-risk volume that no lens can see, then close those with heat, smoke or aspirating detection rather than with another camera. AI Bot Eye's own approach on published deployments is a control-room assessment of back-of-house views first, which is the right sequence: choose the views that carry the risk, not the views that happen to exist.
Two statements sit next to each other. Processing can stay on site, including on isolated recorder networks, without cloud video processing or continuous internet, and SMS and the hooter can work without internet. In the same breath: connected channels such as WhatsApp and cloud dashboards require the relevant network access.
The escalation ladder has two halves with different failure modes, and a site that does not know which is which will design the wrong one as its primary. A remote plant on a flaky link keeps its siren and its SMS during an outage and loses the dashboard image and the WhatsApp thread, which is exactly the pair a night manager was going to rely on to decide whether the alert is real.
Put the channel that has to work on the side of the line that survives, and treat the image-carrying channels as confirmation rather than as notification. Ask for the channel map per site in writing, including which relay outputs are wired and to what, and test the ladder with the uplink pulled rather than with the uplink up.
Nine modules are published: fire and smoke detection, physical intrusion detection, CNC and stack-light monitoring, ANPR, PPE detection, a foot traffic tracker, heatmap insights, camera health monitoring and biometric identification. What each module detects, what it needs from a camera, and how it behaves once tuned are not publicly specified. The documentation covers the module list, not the module.
A count is not coverage. A plant that buys on the strength of nine modules and needs three of them has no published way to tell in advance which of the three is mature and which is a good idea, so the scoping risk lands entirely on the buyer.
Pick the two modules that justify the project and evaluate only those, module by module, with the vendor answering what each needs from a camera view and what a false alert looks like. AI Bot Eye's own demonstration program is the mechanism: prove those two on your own feeds and treat the other seven as a roadmap you have not paid for yet.
The published surface ends at detection and alerting. Retention, where clips or images are stored, export behavior and any case or investigation workflow do not appear on the pages this page read. The documentation covers how an alert reaches somebody, not what the organization holds a month later.
For a site that will need an evidence pack for an insurer, a regulator or an incident report opened weeks after the event, the alerting layer is not the system of record and never claimed to be. The recorder underneath it stays the archive, with whatever retention and export behavior it already has, so a gap in that recorder is not closed by adding intelligence above it.
Ask for retention, storage location and export format in writing before the pilot, and keep the existing recorder as the named system of record in the evidence procedure. Where the recorder's own retention is the real problem, fix that first: an early-warning layer over a recorder that holds seven days will still only prove what seven days can prove.
The same five constraints in one view, sized to paste into an evaluation document.
Swipe the table sideways to see every column.
AI Bot Eye data comes from its own product and solution pages at aiboteye.com, checked on August 17, 2026. Gaps are marked as not publicly specified.
If the cameras and recorders are already in place, the certified fire system is staying, and the board question is how many minutes pass before somebody knows, this product answers exactly that and answers it against recorder brands it names. Its published deployments sit in hotels and plants where an early image on a phone changes the first ten minutes, and a narrow layer that does one job well is often a better purchase than a platform bought for one feature.
Spot AI starts from a different unit of purchase, which is why the constraint list looks different. Fire detection is one of 15+ pre-trained Video AI Agents rather than the product itself, so the same cameras also carry vehicle break-in, after-hours intrusion, personal protective equipment, forklift near-miss, falls and hazard-zone crowding, and Iris builds a detection that is not on the list in natural conversation. The module question becomes which agents you switch on rather than which vendor you add.
The platform is also the system of record. Full-resolution video stays on the Intelligent Video Recorder in the building and only event metadata crosses the network, so retention, search and export sit in one place instead of being inherited from whatever recorder is already there. Response runs through talk down, strobes and horns on standard speakers rather than through a single relay-driven hooter, and any ONVIF or RTSP IP camera connects as it is, with legacy analog coming in through the IVR.
The question is not which product notices a fire sooner, but whether you are buying an alert or the record of what happened next.
None of that makes Spot AI the right answer for every buyer. Spot AI is not a certified life-safety system either, and nothing here should be read as replacing a fire panel. Its fire and vehicle break-in agents shipped through the May 2026 Agents Beta and reach customers on a design-partner or beta basis rather than blanket general availability, it publishes no per-agent accuracy figure any more than AI Bot Eye publishes one per module, and it states compatibility by protocol where AI Bot Eye names eight recorder brands outright. It is also a larger commitment: a team that wants one question answered on the feeds it already has may be better served by a layer than by a platform.
A live pilot on your cameras answers in a week what a spec sheet cannot.
Customer-reported outcomes from named Spot AI customers.
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.
Silver Bay Seafoods replaced fragmented legacy camera systems across 22 locations, including remote Alaska facilities, and lifted operational efficiency 15%.
Liberty-Perry School District resolves an incident in about five minutes, after evaluating ten systems before choosing Spot AI.
"With Spot AI, we're focused on three things: safety, productivity, and security."
Five, all documented on its own site: it is an extra early-warning layer alongside certified fire systems and does not replace certified life-safety systems; detection is bounded by fire becoming visible in a camera view; WhatsApp and the cloud dashboard require network access while on-site processing, SMS and the hooter can work without it; nine modules are named without per-module detail; and retention, storage, export and case workflow are not publicly specified.
Yes, and it names the hardware rather than describing it. The site states it works with feeds from Hikvision, CP Plus, Dahua, Uniview, Prama, Honeywell, Bosch and Hanwha plus most other DVR, NVR and IP systems, and that detection is adapted to each camera view for angle, distance, lighting, activity and known patterns. Processing can stay on site, including on isolated recorder networks.
No, and the constraint is the vendor's own. AI Bot Eye states that it is an additional visual early-warning layer alongside certified fire systems and that it does not replace certified life-safety systems. What is constrained is the compliance role, not the usefulness: an image of a visible fire on a phone in under 5 seconds changes the first ten minutes. Treat it as an extra path to a human.
That is the part not publicly specified. Retention, storage location, export behavior and any case workflow do not appear on the published pages, so the recorder underneath remains the archive with whatever limits it already has. Ask for retention, storage location and export format in writing, and name the recorder as the system of record in your own procedure.
It depends on how many questions the cameras have to answer. If fire early warning is the whole brief and the recorders are staying, a narrow layer on those recorders is a reasonable purchase. If the same feeds also have to carry safety, security and operations, a camera-agnostic platform such as Spot AI fits: 15+ pre-trained Video AI Agents run across the fleet already installed and full-resolution video stays on the Intelligent Video Recorder in the building.