AI Cam

AI Cam is an AI-powered smart surveillance platform. Traditional CCTV simply records — you only find out about an incident when you review the tape afterwards. AI Cam adds intelligence: your cameras are watched in real time, real threats are picked out and independently checked, and each one is written up with a threat level. You get a rich visual alert the moment something matters, instead of hours of footage nobody is watching.

Yes. Open AI Assistant in the AI Cam web sidebar, or AI Cam > AI Assistant in the main IoSnack Android app (version 1.0.114 or later). Both use the same server workflow and account permissions. Camera setup previews permanent settings before Apply. On Android 1.0.115 or later, edited setup hours are previewed again before Apply while keeping your extra AI instructions; a handled AI error can be retried in the same conversation. Temporary watch adds duties to one camera: choose a duration, describe a visible situation, review the preview and press Start. End it when back; Until I return expires after 24 hours at the latest. Choose duration in the controls; a time written in chat does not set expiry. Existing Watchman hours, credits, caps and notification settings still apply, and permitted cloud AI is required. From Android 1.0.116, Ask history searches permitted saved object visits, Watchman logs, events and alert records through the same engine as web Reports, with source counts, coverage and evidence links. The same recorded event may have multiple scans and notifications, so source totals overlap and are not a verified physical incident count. Watchman logs last 7 days and object visits 90 days; missing or expired evidence cannot be recovered. Simple supported questions need no cloud AI; other questions use text interpretation only. Saved records, images and videos are not sent for re-analysis. Updated devices record more model-supported object observations using their existing detection inference, without enabling new alert types. Paid plans allow Assistant by default, with a per-client Allow/Disable control for the platform super-admin. Action permissions and Watchman/Reports plan features still apply. Typed instructions work on web and Android; voice commands are not included. See the AI Cam Help Centre for the full guide.

From AI Cam Box 1.8.37, enabled Activity Reports collect local object observations independently of alert arming and AI Watchman. Collection can continue through Home mode and outside alert arming hours. Camera off, device sleep and backup standby still stop collection, and the local detector must be available. Existing zones and confidence limits apply. This needs no extra cloud AI call, but local processing and metadata uploads use device resources. The Box status line distinguishes activity history from alerts and Watchman; enabled does not prove that an object was detected or uploaded. Phone Box visits are continuous presence per label, not guaranteed individual counts or action recognition. Existing Reports and Assistant access permissions still apply.

No. AI Cam works with the standard HD and IP cameras you already own — anything that provides an RTSP stream. Even if your cameras all run through a DVR or NVR recorder box, nothing gets rewired: the recorder keeps recording and showing on your TV, and AI Cam reads the channels you choose over your own network. You add one small local AI device per site — a mini-PC, a spare computer or even a spare Android phone — and it does the detection. There is no rip-and-replace and no proprietary camera to buy.

Raw motion is never enough to alert you on its own — it has to be a real detection first, and you can draw detection zones so streets, skies and corridors are ignored entirely. Every event that gets through is then independently checked, given a threat level, and false positives are suppressed. Cooldown windows on alert rules then make sure the same event never spams you.

Detection covers people, vehicles, fire and smoke, intrusion, general motion and loitering. Events are then classified across sixteen types, which also include falls, crowds, zone breaches, packages, animals, weapons, bags, phones and screens. Beyond any fixed list, the AI Watchman judges the scene itself — so a danger nobody wrote a rule for is still reported. Each camera can have its own detection zones and its own detection on/off toggle, so coverage is tuned exactly to each location.

An alert rule answers a fixed question: is there a person, a vehicle, a fire? The AI Watchman asks an open one — is anything here wrong? It watches the way a night guard would and reports what it finds, including dangers nobody configured a rule for: someone who has fallen and is not moving, a fight, a weapon, a snake in the yard, a child alone near water, smoke already filling a room. You can give it standing instructions in plain English, such as "watch the kids near the pool" — those are duties added to its watch, never a filter on what else it may report. Findings are rated from low to critical and you set the level that reaches you, with a daily limit and a cooldown you choose so the cost stays predictable.

Alerts arrive by email and push notification with an event thumbnail and an AI-written description. You build alert rules around event type, priority, device, camera and schedule — including business-hours awareness — and the dashboard also gives immediate in-browser audio-visual warnings for critical events.