The Platform

AIS

An on-premises security appliance that watches a property continuously, explains what it sees in plain language, and keeps every frame of it on hardware you own.

How it works

Four layers, one event stream

The reason a normal camera system feels dumb is that every part of it lives alone. AIS is built the other way around: everything the property senses lands in one place, and everything you see is a read of that one place.

01

Ingest and detect

Cameras arrive over PoE into an NVR core with a dedicated inference accelerator. Object, face, and plate detection run at the edge, on the appliance, at full frame rate.

02

Sense and fuse

Z-Wave and Zigbee door, window, motion, and glass-break sensors join the same event bus as the cameras. Sensor freshness is tracked explicitly — a sensor that has gone quiet reads as unknown, never as safe.

03

Interpret

A local vision model describes scenes in language. A four-dimensional alert matrix — who, where, when, what — decides whether something is worth your attention, with cooldowns and learned false-alarm suppression.

04

Surface

One dashboard, one mobile app, one push channel. Live view, events, identity, incidents, automations, and the alarm panel, all reading the same stream.

Sovereignty

The data plane stays home. Only the control plane travels.

Managing appliances across properties needs some central signal — health, updates, diagnostics. Rather than describe that in a privacy policy, we draw the line and publish it. Here is precisely what crosses, in both directions, and what cannot.

Your property

The appliance in your building

Never crosses the line

  • ✕Video and recordings
  • ✕Face and plate images
  • ✕Scene descriptions
  • ✕Any personal or biometric data

Recording, recognition, and scene analysis all resolve here. There is no second copy anywhere, because one is never made.

Our control plane

Fleet health only

Leaves the property →

  • →Heartbeat and subsystem status
  • →Camera online state and frame rate
  • →Storage, CPU, GPU, temperature
  • →Error events (metadata only)
  • →Appliance version

← Comes back

  • ←Configuration changes
  • ←Update packages
  • ←Restart and self-heal commands
  • ←Remote-diagnostic session grants

Compromise our infrastructure completely and you get uptime figures, temperatures, and version numbers. Not one frame of anyone's video.

Against the alternatives

Enterprise-grade detection without the enterprise cloud

Verkada and Avigilon build good products, and two rows below go to them. A comparison that flatters only its author tells you nothing — this one is built so it survives your due diligence rather than collapsing during it.

Where footage lives

Our advantage

Cloud VMS

Uploaded to the vendor's cloud. Their staff, their access controls, their breach surface.

AIS

On the appliance in your building. There is no second copy to compromise, because one was never made.

Where faces are matched

Our advantage

Cloud VMS

Face templates and identity data typically leave the site for vendor infrastructure.

AIS

Face and plate recognition run on your hardware. Biometric data never crosses the boundary.

If the vendor is breached

Our advantage

Cloud VMS

Interior footage from customer sites has been reached this way before. It is not a hypothetical.

AIS

A total compromise of our infrastructure exposes fleet health metadata — uptime, temperature, version numbers.

Network security

Our advantage

Cloud VMS

Cameras phone home continuously. The camera VLAN is usually the customer's problem.

AIS

Cameras are default-denied to the internet. The optional hardening module adds firewall, IDS, SIEM, and endpoint telemetry as one deployment.

Support and SLA

Their advantage

Cloud VMS

24/7 support organisations, hardware advance-replacement, contractual uptime guarantees.

AIS

The engineers who built your system answer the phone. No tier one, no ticket number, no contractual uptime figure — if you need an SLA on paper, theirs is the stronger offer.

Scale and installer network

Their advantage

Cloud VMS

Thousands of certified integrators. A site anywhere, provisioned quickly.

AIS

A vetted integrator channel building to our specification. Deliberately narrow — we would rather engineer every deployment than certify a partner and hope. If you need forty sites live this quarter, they can do that and we cannot.

Updates and maintenance

Even

Cloud VMS

Pushed automatically. You never think about it — and you never control it either.

AIS

Staged updates you approve. Nothing changes on your property on a vendor's release schedule.

AI detection quality

Even

Cloud VMS

Mature models improved by a decade of footage from every customer site they hold.

AIS

Native face and plate recognition with local scene analysis, tuned against your property rather than an aggregate of everyone else's. The trade is real: their models have seen more, and your footage was never part of the training set.

Capabilities

Everything the platform does

Each capability carries its true state. In production is running on a live deployment today. In the platform is real, running code we present as an aid rather than a guarantee. On the roadmap is specified and not yet built.

In production

Face and Plate Recognition

Live cameras with face and licence plate recognition running natively on the appliance. Plate matching is deliberately strict — a garbled read resolves to unknown rather than silently matching a household vehicle.

In production

Signed Incident Evidence Packets

An incident exports as an ed25519-signed ZIP: PDF summary, snapshot, clip, and manifest. A bundled command-line verifier checks the signature independently of us.

In production

Continuous AI-Assisted Scene Analysis

A local vision model describes what it sees, in language, so events read as sentences rather than bounding boxes. Best-effort — the model runs on your hardware, not ours.

In production

Two-Way Audio at Gates and Doors

Talk down to a delivery driver or a stranger from your phone, from anywhere you have access to the system.

In production

Z-Wave and Zigbee Sensor Fusion

Door, window, motion, and glass-break sensors on the same timeline as the cameras — with freshness tracking, so a stale sensor reads as stale rather than as safe.

In production

Alarm Panel with Duress Codes

Per-user PINs, an audit-logged duress flow, and hold-to-fire panic from any screen in the app.

In production

Context-Aware Alerting

A four-dimensional alert matrix — who, where, when, what — with cooldowns and learned false-alarm suppression. Unknown people, watchlist hits, and loitering can never be suppressed.

In production

Mobile and Push

Native Android app plus an installable web app, with push notifications through FCM and Web Push.

In the platform

Weapon Detection Pipeline

A trained detector with per-camera false-positive learning and adaptive suppression, so the operator tunes it rather than drowning in it. Probabilistic — we treat it as an aid to attention, not a guarantee.

In the platform

Automation Rules Engine

Conditions, schedules, and actions — lights, siren, notifications — built from the same event stream the dashboard reads.

In the platform

Spatial View

Cameras and sensors pinned to a floorplan, so an event has a place and not just a camera name.

On the roadmap

Fleet Control Plane

A thin cloud plane for health monitoring, staged over-the-air updates, and remote diagnostics across many sites. Metadata only — never a frame of video.

Evidence

An incident leaves the building as a sealed archive

Most systems hand your insurer a clip off a DVR that anyone could have edited. An AIS incident exports as a signed packet that your own advisor verifies, offline, on their machine.

What exports

incident-packet.zip

  • report.pdf

    Human-readable incident report — what happened, when, which camera

  • thumbnail.jpg

    Snapshot from the camera at the moment of detection

  • clip.mp4

    The recorded video clip around the event

  • manifest.json

    Every file above with its SHA-256 hash, plus event metadata

  • manifest.sig

    Ed25519 signature over the manifest, made by this installation's key

  • public_key.pem

    The installation's public key, so the signature can be checked offline

Verified without us

$ ais-verify-packet incident-packet.zip

OK — hashes verified, signature valid

The verifier is open source and runs offline. Your counsel, your insurer, or their forensic examiner runs it on their own machine — no account, no portal, and no request to us. The packet carries the signer fingerprint your installer recorded at commissioning, so authenticity is checked against a record you already hold.

Alter one frame of the clip and the hash stops matching. Re-sign the manifest and the key no longer matches the appliance that recorded it.

What this is not: a notarised timestamp. The signature proves the packet is unaltered since your appliance sealed it and that this appliance sealed it — not what the clock read to a third party. That link is the next one we are adding.

The interface

It opens on an answer, not a grid of cameras

The first thing the app shows is a sentence: all quiet, or the thing that is not. Underneath that sits the arming panel, the system's own suggestions for tuning itself, and the live wall — in that order, because that is the order the questions actually arrive in.

  • The system proposes its own tuning — which camera is crying wolf, and what to do about it.
  • Hold-to-fire panic is reachable from every screen.
  • Events read as narrative, not as bounding-box coordinates.
  • Native Android app plus an installable web app, with push on both.

Specified for the property, not shipped from a catalogue.

Sized against the survey — buildings, approaches, sightlines, and what is already on site.

The appliance

Real hardware, specified for the property

There is no version of this that runs well on an under-specified box in a cupboard. Every deployment starts with a site survey and a build sheet sized to the camera count, the buildings, and the way the property is actually laid out.

Compute

Dedicated NVR-class server with GPU decode and inference

Storage

Surveillance-rated RAID array, ~45 days continuous at 20 cameras

Power

Pure sine UPS for clean shutdown on outage

Network

PoE switching, fibre between buildings, isolated camera VLAN

Remote sites

Wireless-linked buildings record locally so a dropped or jammed link cannot cost you footage

Access

Private mesh network — no port forwarding, no exposed NVR

Editions

Choose the surface, not a different system

Both editions run the same appliance, the same detection, and the same event model. The edition decides what the interface leads with and which surfaces exist at all.

AIS Estate

AI-Augmented Estate Intelligence

Principals, family offices, and private estates

One property, or several, held under a single calm surface. Estate answers the only question a principal actually asks — is everything alright — and keeps the rest out of the way until it matters.

  • Narrative daily briefing instead of an alert firehose
  • Watchlist by face and plate
  • Travel mode and scheduled privacy blackout zones
  • Concierge line to the people who built the system
  • Household roles: Principal, Family, Staff, Detail
Explore AIS Estate

AIS Command

AI-Augmented Site Operations

Multi-site operators — schools, retail, offices, facilities

The same engine pointed at operations rather than a household. Sites, buildings, and floors in one hierarchy, with the triage queue, case workflow, and audit trail a real security team needs.

  • Site → building → floor hierarchy with scoped data
  • Operator video wall and alarm-driven triage queue
  • Investigations with full case workflow
  • Role-based access control and audit log explorer
  • SSO (SAML / OIDC) and webhook / SIEM integration
Explore AIS Command

Add-on module

AIS Bastion

On-Site Network Hardening and Local Threat Detection

A camera system is only as private as the network it sits on. Bastion is the other half of the estate: a hardened firewall, intrusion detection, a full SIEM, endpoint telemetry, and an AI analyst — all running on site, with nothing shipped to a cloud console.

Explore AIS Bastion
Firewall and intrusion detection
SIEM
Endpoint telemetry
Local AI analyst
Network visibility and live map
Camera and IoT isolation

Due diligence

Four claims worth challenging in any security proposal

Ours included. Each of these is standard in this industry and none survives a technical question. Here is the question to ask, and where we land on it.

“Guaranteed break-in detection”

No camera system can promise this, and any vendor who does is selling you a camera angle, a hood, and bad light. Ask instead what can never be suppressed. Ours: unknown people, watchlist matches, and loitering are structurally excluded from the tuning logic, at any configuration.

“Courtroom-ready evidence”

Ask how the file is signed and who can verify it. Ours export as ed25519-signed packets your counsel verifies with a tool that does not depend on us — most systems produce an unsigned file off a DVR. Notarised timestamping is the next link in that chain, and until it ships we use the narrower phrase.

“24/7 AI monitoring”

Ask who is awake. Almost always the answer is nobody, and the phrase means software was running. Ours runs on your hardware, best-effort, with no operations centre and no on-call human behind it. We say AI-augmented because that is the literal description.

“Tamper-proof cameras”

Nothing bolted to an exterior wall is tamper-proof. Ask what the system notices. Ours detects a dropped stream today; scene change, lens obstruction, and PoE loss are specified and not yet shipped.

Questions

Asked and answered

No. Recording, detection, face and plate recognition, and scene analysis all run on the appliance in your building. The system is designed to run fully standalone — if the internet drops, it keeps recording, keeps alarming, and stays reachable on your own network.

Every property is different. Let us look at yours.

A briefing is a conversation about the property, the threat model, and what already exists on site. If the honest answer is that your current system is adequate, we will tell you that.

Request a Briefing