Platform

Detect, report, act.

The platform is where the estate is run from. It receives, analyses and quantifies every image against the criteria you set, alerts only when something warrants it, and lets you manage a thousand cameras as easily as one.

Alert rules

Say what worries you. The platform makes it a rule.

You describe the condition in plain language. The platform turns that into a prompt with named, quantitative outputs the model must return for every frame, and builds boolean rules on those outputs that decide whether to alert, how urgently, and who is told. Every step is shown to you and can be edited.

New alert rule · Plant room compound
1You write
“Tell me if anyone is in the plant room compound outside working hours, and whether they are wearing hi-vis.”

A sentence, the way you would brief a colleague.

2The platform derives
Prompt sent with every escalated frame

Report the people visible inside the fenced compound and whether each wears a hi-vis vest. Return the fields below as values, not prose.

Named outputs
people_insidenumber
all_hi_visyes / no
vehicle_presentyes / no
activityshort text
confidence0 to 1

Structured answers, so the next step can be arithmetic rather than interpretation.

3Rules decide
Whetherpeople_inside > 0 AND NOT all_hi_vis AND time 18:00–06:00 AND confidence ≥ 0.7
HowPush and SMS with the frame and activity. Hold-off 20 min, re-arm when the compound is clear, acknowledgement required.
WhoSecurity duty officer; site manager if not acknowledged in 10 min; webhook to the control-room log.

Boolean rules on the outputs, with sensor readings and time of day available as extra terms.

The rule is stored with all three parts. Edit the sentence and the platform re-derives the rest; edit an output or a rule and it keeps your change.
  • Quantitative, not vague. Counts, levels, yes/no answers and a confidence come back as values you can threshold, chart and audit.
  • And, or, not. Outputs combine with each other, with temperature, humidity and light, and with time of day.
  • Quiet by design. Hold-off stops a busy scene sending fifty alerts; re-arm means you hear again only after the scene has cleared; acknowledgement is for alerts a person must see.
From image to action

What happens to a frame.

  1. Ingest

    Every image lands in your tenant

    Authenticated, rate-limited upload from each camera. Encrypted in transit and at rest. Battery, light, temperature and signal arrive with every frame and as their own time series.

  2. Analyse

    Read against your criteria

    A vision-language model describes the scene, tags what it sees, and answers the questions you wrote: a count, a yes or no, a level, a description of who and what.

  3. Decide

    Rules, not feeds

    Rules combine what the AI found with sensor readings using and, or and not. Hold-off, re-arm and acknowledgement keep repeat alerts quiet.

  4. Tell

    Reach the right person

    Push, SMS, email or webhook, with the image and a readable account. Everything else becomes searchable history you can graph or reprocess when models improve.

Estate
  • Northern Water142 cameras
    • Rivers38 · Flood profile
      • Ford Bridge ● online · 97%
      • Mill Weir ● online · 88%
      • Low Road ● online · 31%
    • Substations61 · Intrusion profile
    • Pumping stations43 · Intrusion profile
Customer, department, site, camera. Configuration applies at any level.
Fleet

Deploying the thousandth camera is no harder than the first.

Cameras sit in a tree of customer, department, site and camera. Rules, capture schedules and power policy apply at whatever level you choose and push to every camera below it over the air.

  • Batch configuration: change a group, not a camera.
  • Live health: battery, signal, last seen, and predicted service date for every unit.
  • Firmware over the air with rollback if a camera fails to boot.
  • SIMs, data and billing inside the service, so nothing reaches procurement or finance.
  • Deployment by hand: position it, press the button, it appears in the tree within minutes.
Governance

Built for organisations with compliance obligations.

Transport and storageTLS everywhere; images and metadata encrypted at rest in your own tenant.
AccessHierarchical, role-based access control with multi-factor authentication.
AuditEvery login, configuration change and acknowledgement is logged.
RetentionTwelve months by default, configurable per customer, with export and deletion on request.
Data protectionGDPR-aligned processing; images are analysed to answer the criteria you set and nothing else.
HostingUK and EU hosting; cross-tenant isolation by design.
Integration

Alerts and imagery reach your existing systems.

Webhooks fire on every alert with the image, the AI reading and the camera's metadata. The API exposes cameras, images, analyses, alerts and sensor series. Export takes the rest.

POST https://ops.example.com/hooks/machaira
{
  "event": "alert.triggered",
  "rule": "Plant room after hours",
  "camera": { "id": 4821, "name": "Plant room east gate" },
  "triggeredAt": "2026-08-20T22:16:41Z",
  "summary": "2 people entered plant room compound, both on foot, no vehicle present, approached from the eastern gate.",
  "image": "https://…/images/918273.jpg",
  "analysis": { "people": 2, "vehicle": false, "hi_vis": false },
  "sensors": { "temperature": 12.4, "light": 3 }
}

See the portal on your own scene.

We set up a tenant, a camera and your criteria, and you watch the first alerts arrive.

Request a demo