What it senses
Omaya reads the signals your building already emits, plus whatever you choose to add. Nothing here requires an app on anyone's phone.
People and assets
Anonymous by default; identified only where you deliberately bind a tag to a person or asset.
-
Bluetooth tags — staff, contractors, trolleys, equipment
-
Radar and presence sensors — headcount in a room and where in it, no identity
-
Door counters — entries and exits at a threshold
-
Unrecognised devices — counted as passing footfall, never identified
-
GPS trackers — outdoors, over LoRaWAN or their own cellular link
-
Badges with their own SIM — GPS outdoors, the beacons or Wi-Fi access points they hear indoors, and SOS and man-down alarms
-
Falls and man-down — from the accelerometer in a worn tag (preview)
Environment
Conditions that decide whether a space is usable, comfortable and compliant.
-
Temperature and humidity
-
Noise level
-
Air quality
-
Leak and supply-level sensors
Building systems
The equipment already installed, brought into the same picture as everything else.
-
Energy meters, down to a single circuit
-
Data-hall plant — UPS units, rack PDUs and inlet sensors, cooling units, fan walls and generators — as readings an edge gateway has already taken
-
Switches, actuators, lights, air-conditioning and blinds — read and controlled through the smart-home bridge
-
Buttons, panic buttons and call points — a panic press reaches the people named for it by email, SMS, WhatsApp or push, and says where the person is
-
Gateway health, battery and signal strength
-
Room bookings from your Google or Microsoft 365 calendars
Signals arrive over Bluetooth, Wi-Fi, Matter and LoRaWAN, and over wired links where the equipment provides one. Device families are decoded individually rather than reduced to a lowest common denominator, so a counter reports counts and a meter reports consumption.
What it works out
Raw signal is not an answer. Between the sensor and the report, Omaya resolves where something is, how long it stayed and what that means for the space.
Position
Methods chosen per venue by what the site can actually support.
-
Presence — which zone heard it best
-
Triangulation — a position from three or more anchors
-
Fingerprinting — matched against a surveyed signal map
-
Anchors — a badge that listens rather than broadcasts, placed by the beacons or Wi-Fi access points it hears, each registered to a room
-
GPS outdoors — a geofence visit, a trip or a stop, handed back to the room as soon as a gateway hears the tag indoors
Stability
A device on a boundary should not flicker between two rooms all afternoon.
-
Smoothing across successive readings
-
Hysteresis before a zone change is committed
-
A floor model that weighs the evidence over time, so a floor below cannot claim a device
-
A gateway that goes quiet for a minute keeps its tags until a neighbour has won twice
-
Stairs, lifts and escalators modelled, so a path between floors is a real path
-
One zone per device per period, never counted twice
-
Radar checked against a known tag, so the coverage of a sensor is measured, not assumed
Meaning
Derived measures, computed continuously rather than at report time.
-
Dwell — how long a stay lasted, bucketed
-
Occupancy against a zone's capacity
-
People inside from door counters — entries less exits, reset each day
-
Return behaviour — first seen versus coming back
-
Forecast occupancy from observed pattern
-
Habit — where each person usually is, and when each room is usually used
-
Anomalies — a device that broke its own reporting pattern, scored and deduplicated
What it reports
27 built-in reports in the Reports menu, plus the live views and each app's own, each answering one question. Grouped here by the question rather than by where they sit in the menu.
Who was here, and where
-
Footfall — how many were in the whole organisation, a location, a venue or a zone each day or hour: the tags people carry, counted once per place however many rooms they went through, with tags on assets counted apart, beside the door counters' entries, the most inside at once against capacity, and how long tags stayed; against the period before or last year, and as a typical week
-
Attendance — per person and day: first in, last out, time on site with room stays joined up, and time on each floor; days present and the attendance rate against the days counted; late arrivals and early leavers against opening hours; and who is not in yet today
-
Asset utilisation — how much each trolley, wheelchair or forklift was used and where it stood; the idle ones, and the missing ones with where each was last seen
-
Detection heatmap — devices per zone, venue or location and day, shaded
-
Zone presence — where time was actually spent
-
Sensor trail — zone-by-zone history for one device or entity
-
People count — live headcount, from counters, radar or tags
-
Room usage map — where in a room people actually stand, from radar
-
Movement flow — which zones feed which, and in what order — within a floor or between floors, with the most common routes through a visit
-
Enter / exit — crossings at a threshold, and how many people are inside now and at peak
-
Outdoor activity — time outside, distance, trips, stops and geofence visits per GPS tracker and per geofence
-
Gender and entry — where a counter reports it
How long, and how often
-
Dwell time — stays bucketed by duration
-
Device returns — first seen versus returning, by interval
-
Last 24 hours — the day so far, live
-
Distance — separation between devices over time
-
Zone compare — two spaces side by side, or one place against an earlier period
-
Space utilisation — every zone against its capacity, ranked
-
Impact analysis — what changed after an event you mark
Is the estate healthy
-
Device status — online, stale, offline; how well each device is heard, from strong to weak; and the last values it sent
-
Battery — level and projected replacement
-
Environment — conditions hour by hour against the weather outside, threshold breaches, and a compliance check of every room against a temperature range, all day or during opening hours only
-
Button press — call points and request buttons
-
Alerts — what fired, when, and what followed
-
Rule & alert effectiveness — how often each rule fires and whether its actions went out; how quickly alerts are acknowledged; rules that could not be checked for lack of data
-
Panic & incident response — how fast guards picked up and resolved incidents and panic presses, false alarms, each guard's workload, and the presses nobody answered
-
Fall & man-down (preview) — what the fall detector flagged, how much was real once events are labelled, and how fast people answered
-
Gateway availability — each gateway's uptime against an SLA target, its outages and an hour-by-hour timeline
Prove it, and take it away
-
Compliance — evidence against a required standard
-
Meeting room — booked against actually used
-
Custom report builder — pick the table, columns and filters
-
Ask in plain language — describe the report, get the table and chart
-
Export — scheduled delivery to inbox or storage, with the report attached: footfall, attendance, entries and exits, people count, dwell, assets, gateways, incidents, and a monthly site pack covering the whole platform on one PDF
-
Scheduled by the app's own people — whoever runs Facility, Toilet Hygiene, Visitors, Energy, Security Watch, the warehouse or the data hall can schedule that app's reports from inside it, to its own users, holding only what they may see
Every report filters by location, venue, zone and date range (the people reports by label too), converts timestamps to the site's own timezone, and exports to CSV. Per-day figures can count the site's operating days or every calendar day, and a day the site sent no data is marked as missing rather than counted as empty. Readings that could not be placed in a room are attributed to their venue rather than dropped, so a total never quietly loses rows.
What it does without being asked
A rule watches a measure or an event, decides whether a condition holds, and acts. 35 can be watched, joined with AND / OR left to right, with time-of-day and day-of-week windows, for a zone, a device, a label, a person or asset, or a gateway.
What a rule can watch
-
People count, enter count, exit count, zone density
-
Dwell time, home-zone status
-
Temperature, humidity, noise, air quality
-
Distance between devices
-
Battery level, device status, signal strength
-
Radar: people in a zone, minutes occupied, minutes empty
-
Zone utilisation and hours at capacity over the last week
-
Power and energy today, net of on-site generation
-
Visitor events, and a GPS tracker entering or leaving a geofence
-
Button pressed, any field a sensor's decoder produces, and any custom field you define
-
Someone entering or leaving a zone, venue or location; a gateway going down or coming back
-
An anomaly, a work order opened or resolved, a Security Watch incident, a fall or man-down
-
A date in a person's or asset's own fields, as days from today
What it can do about it
-
Notify by email, SMS, WhatsApp or Telegram
-
Raise a job in your maintenance system
-
Call a webhook, with retry and a dead-letter trail
-
Switch lights, air-conditioning, blinds or any device the smart-home bridge exposes — any service, any entity, targeted by the rule that fired
-
Locate a badge, make it sound, or end its alarm
-
Escalate if nobody acknowledges
-
Every firing logged — what each condition saw and what each action did — with a tag rule's cooldown held per device
Rules can be tested against up to 30 days of real history before you save them, so you find out what a threshold would have done last month rather than next month. The assistant's daily review proposes rules too — you accept one and it opens pre-filled in the rule editor.
What it thinks about
The assistant works from what the platform already knows — every zone, device, rule and reading of one organisation — and is built to explain and propose rather than to act on its own.
OAIA — the operations assistant
Switched on per organisation from the A.I Model card; off by default.
-
Heartbeat every 2, 4, 6, 8, 12 or 24 hours: gateways offline or weak, ingestion gone silent, batteries, devices off their pattern, environment off baseline
-
Speaks only on change — new, escalated or cleared — and tells every administrator
-
Daily review (1, 2, 3 or 7 days): proposes rules, safety and comfort improvements, configuration fixes — conditioned on sense, such as occupancy before comfort
-
Findings acknowledged, snoozed, reopened or marked won't-fix; cleared ones hidden but kept
-
Chat that knows the tenant, the clock and the asker's role, shows each step as it works, and links the report or device behind an answer; long conversations are summarised automatically, and /compact and /new are still there
-
Ask on the web, in the mobile app, or over Telegram in a private chat with the organisation's bot; a change proposed away from the web is applied on the web
-
Ready-made lookups answer the common questions in one step — gateway status, a device and what it is reading now, who is in a zone, what needs attention
-
A thumbs up or down on every answer; the most-asked questions and the answers marked wrong are listed for the platform team
-
Approved actions it can draft — venue hours, zone capacity, an alert rule you ask for, a rule on or off, a device's zone, a gateway's receive offset measured on the Calibration Bench, finding and suggestion states, plus each app's own — applied only by a person with the permission, once
-
Checks its own work: after a change is applied, it looks again later and says whether it fixed what it was for
-
Suggests follow-up questions after each answer, and once a day rechecks older suggestions and closes the ones whose problem has gone
Autopilot (beta)
The only way the assistant ever changes anything without a person clicking Apply. Off by default.
-
Three switches: the platform makes it available per organisation, the organisation picks Off / Shadow / Act, and turns on tiers — Draft (create proposed rules switched off), Configure (a capacity, a venue's hours, an unassigned device's zone), Operate (cooling on for a hot, occupied room; fans for stale air; lights for people in the dark or a zone with an open critical incident; everything off after the radar says a room has been empty ten minutes or the venue has been closed half an hour with nobody counted)
-
Limits in code: three actions per heartbeat, twenty a day, an hour before touching the same thing again, never flipping a device back within thirty minutes, setpoints inside 20–28 °C; every action listed with Undo; an undo pauses that kind for a day, three in a week drop it to Shadow
-
Its list is fixed and reversible; none of the apps' actions above is on it, however the switches are set
-
Switching it to Act, or making it available to an organisation, asks for confirmation that it is beta and can get one wrong
One assistant, every app
Each app you switch on plugs into the assistant; an app you have not switched on is never mentioned.
-
Hears: security incidents and judgements; bookings beside the hours a room was really used; work orders, assets and their service dates; every toilet's score, cleans and SLA breaches; kilowatt-hours beside occupancy and after-hours draw; connector health and webhook deliveries; visits and hosts; compliance verdicts and calibrations; drills and who was missing; a data hall's PUE, load split, open alerts and operating-practice checks; warehouse stock, docks and gate passes; where a tracker went outdoors
-
Says: each app's week goes into the daily review with its own gaps, and its stalls become heartbeat findings — an incident nobody acknowledged, a calendar that stopped syncing, an urgent order nobody took, a dirty toilet with nobody sent, a bridge that stopped answering, a drill overdue, a data-hall alert nobody acknowledged or plant that went quiet
-
Proposes: whitelist a person or set armed hours; release a no-show booking, tune automatic release, make a room bookable; open, re-prioritise or cancel a work order; dispatch a clean, make a room a toilet, set a toilet's target or trigger; re-send or pause a webhook; run a compliance assessment or switch a profile; set a data hall's inlet limits, or the site's PUE target and tariff; set a warehouse's QC sampling or open a count — all approve-then-apply
-
Never: closes an incident, books or cancels a meeting, marks work or a clean done, edits a standard, touches a shift, changes an endpoint or a key, starts, ends or marks a roll call, or touches a data-hall alert, rack or device
-
Every app's settings page carries Ask OAIA, and a record — an incident, a booking, a work order, a toilet, a verdict, a drill — can be put in front of the chat from the page it lives on
Security judgement
Inside the Security Watch app, for the guard on duty.
-
Every new incident gets a short assessment and a suggested first step from a fixed list
-
Out of habit: a person somewhere they have never been, or a room in use at an hour it never is and no booking — judged against 30 days of habit, so nobody maintains a list of who may enter where
-
Only an "abnormal" verdict becomes an incident; every verdict is recorded and shown to the administrator
-
Detection itself is deterministic and runs without the model
Where the fence is
The parts that let a customer's security team say yes.
-
One organisation per assistant, through read-only views with the organisation baked in
-
A database account that can only read those views, capped at 200 rows and five seconds
-
Secrets and credentials redacted before the model sees a row
-
No shell, no file system, no other tenant — by construction, then tested with injection, jailbreak and cross-tenant requests
-
Bring your own model key with no cap, or use Omaya's service within a monthly allowance set per organisation
-
The platform can withhold the assistant from an organisation; the organisation can switch it off
What it connects to
Bringing data in
Gateways and bridges speak to Omaya over whichever transport they support.
-
Message-broker, HTTP, TCP and UDP gateway endpoints
-
Smart-home bridge: read any entity it knows, and switch lights, climate and covers back through it
-
Gateways from Dusun, MOKO, Milesight, MikroTik and Minew out of the box, and any gateway that can publish MQTT in the documented format
-
Auto-discovery of devices a gateway starts hearing
-
LoRaWAN through the network server — ChirpStack, or the one built into Milesight gateways — including ChirpStack's standard Cayenne LPP codec
-
Already-decoded readings from an edge gateway that polls plant over Modbus, OPC UA or SNMP — power, energy, voltage, current, temperature, fan and airflow
-
Google and Microsoft 365 calendars for room bookings
-
Your ERP: items, partners and open orders pulled each night, with stock movements sent back
-
GPS trackers that report their own position over a cellular link
-
Cellular badges that report over the message broker directly, and take settings and commands back
-
CAD floor plans as DXF: walls, rooms and doors taken from the drawing
-
Your identity provider for administrator sign-in, with group-to-role mapping
Sending it out
Alerts, exports and events, to where your team already works.
-
Email, SMS, WhatsApp, Telegram
-
Outbound webhooks with signed payloads — including a tracker entering or leaving a geofence
-
External maintenance and CMMS systems
-
Object storage for scheduled report delivery
-
Wall-mounted dashboards for a floor or lobby, including a live 3D floor plan
-
A public API with interactive documentation, including GPS whereabouts, points and geofences
Ready-made for a specific job
The platform underneath is the same. These are pre-built for problems common enough to be worth solving once.
Meeting rooms
Booked against actually occupied — find the rooms held and never used, and free them.
Washroom hygiene
Cleaning triggered by traffic rather than by clock, with supply levels and leak detection. A schedule board shows who cleaned, when and for how long, and marks the jobs that ran late against a deadline that allows for the clean itself. Each toilet has a main and backup cleaner, and can choose who is told about a job, step by step, with a preview of every notification.
Energy
Consumption per circuit against occupancy, so waste is visible as waste. Sub-meters count once under the meter that feeds them, solar and generators offset the total, and cost comes from your rate per kWh. Energy per person, and how much is used while nobody is there.
Facility requests
Button and app requests routed to whoever handles them, tracked to completion. Omaya raises jobs of its own too — a critical battery (a charge job for a rechargeable device), a silent device, an offline gateway, a silent room counter — each of which can be switched off.
Lobby dashboards
A screen showing live occupancy, air quality or availability, without a login — including a live 3D floor plan, and a widget for each app you have switched on. Each board's share link can expire after a set number of days.
Roll call
Emergency muster: a snapshot of everyone on site, who has reached the gathering point, and who is missing. Someone outside on GPS counts as off-site, not missing, and a muster can gather at an outdoor point, a geofence, where a GPS tag inside it counts as arrived.
Visitors
Registration, check-in and check-out, host notification, and a visitor app the organisation brands as its own.
Compliance
Evidence against a required standard, collected as it happens rather than reconstructed later.
Security Watch
For the security team: armed hours, restricted zones, panic buttons, heat that looks like fire, coverage lost, and out-of-habit presence judged by AI. Other apps raise incidents too, such as goods overdue on a gate pass. Guards get a phone app with a one-glance board; escalation goes to named people by email, SMS or WhatsApp. A Reports tab shows supervisors how fast incidents were picked up and resolved, and which were genuine, false alarms, tests or duplicates.
Data centres
A data hall or comms room watched like the rest of the building: rack inlet heat, the power split, and the plant that keeps it running. Readings arrive from an edge gateway; each device is given a role once, in Setup. With a dark theme for the wall screen.
-
Power and PUE — facility, IT, cooling and other load live; each side falls back to what the site actually meters (IT meter, else the rack PDUs), and the card says which basis is in use rather than showing a clean number that is quietly estimated
-
The plant — UPS on battery, battery low and load; cooling units and fan walls stopped or supplying warm air; a fan wall's running, failed and installed fans, airflow and pressure; a generator that has not started while a UPS is on battery. A unit can be marked standby or in maintenance so it raises nothing it should not
-
Alerts that hold their nerve — one per subject, named for the worst rack; escalates past the critical limit or while still climbing, never steps down while open, clears only once settled — and a unit that goes quiet keeps its alerts and gets a "not reporting" one of its own
-
Floor plan in 2D and 3D — racks coloured by inlet temperature and every plant unit drawn at its real size and angle, recoloured live by the alert it carries
-
Operating practice — 17 checks against recognised standards (thermal envelope, N+1 on cooling and UPS, fan-wall headroom, the 80 % rack-load rule, monthly generator test, PUE measurement category, sensor coverage, time to acknowledge) on the site's own readings; where it lacks what a check needs it says not measured rather than guessing
-
AI review — one call on the organisation's own AI allowance picks the two or three unmet checks to do first, reasoning about what is actually at risk here; any check opens straight into the OAIA chat
-
Trends — daily energy by IT, cooling and other, PUE against baseline and target, cost at your tariff, and a log of every time a UPS went on battery or a generator ran
-
Report — any period: time each rack spent in the recommended, allowable and out bands, PUE from metered energy, UPS time on battery, alert response and rack power against rating
Warehouses
A warehouse run on scans, from a tablet on the floor, a desk or the phone. Every box is a labelled unit, and every change is one line of a ledger that is only ever added to.
-
In — receiving with QC sampling, putaway with bin suggestions and storage-class guards
-
Out — picking with FEFO / FIFO reservation and a guided next unit, outbound QC, packing and shipping
-
In between — counts with approved variances, transfers, quarantine, work orders with material issue and returns, gate passes, returns checked against what was shipped, and return to supplier
-
ERP in step — orders and items in by API, CSV or a nightly pull; every movement back out as a webhook or a feed; printed goods received notes, delivery orders and gate passes
-
What the building adds — tagged pallets and forklifts on the live map and flagged when a tag disagrees with its bin; cold-chain exposure per unit; dock occupancy and dwell; goods overdue on a gate pass or moving after hours raised to Security Watch
-
Reports — orders, on-time %, cycle time and backlog; picks in full and per picker per hour; slow movers, aging, stock-outs and bin use; count accuracy — and not recorded where the app keeps no record, never 0
Fall detection (preview)
Falls and man-down from worn tags, with no calibration. The person is asked "Are you OK?"; without an answer it becomes a critical incident in Security Watch for the guards. It only records until alerts are switched on.
How it runs
Separation
-
Each customer's data in its own store, not a shared table with a filter
-
Roles and per-page permissions; a role can be limited to the people and assets carrying chosen labels
-
Every administrative action recorded in an audit trail
-
Retention windows you set, enforced automatically
-
Single sign-on per organisation; direct login can be switched off
-
Two-factor and password expiry enforced on every admin page; app sign-in locks out one person, not the whole office
-
A usage page per organisation: AI allowance, SMS, WhatsApp, email, webhooks, alerts — with failures listed
Keeping itself current
-
Scheduled background tasks, from ingestion to report rollups
-
Gateway health watched continuously, offline raised as an alert
-
Forecasts and aggregates rebuilt on their own schedule
-
Failed outbound deliveries retried, then surfaced rather than lost
Reach
-
Interface in English and Malay; the apps also in Simplified Chinese; five further locales in progress
-
Right-to-left layouts supported
-
Each site keeps its own timezone; reports convert to it
-
Works on a phone, a desk browser and a wall screen