When a freezer fails at 2am, the alarm has to be right. Elpis is built to be right — and to say so when it does not know.
Cold rooms, freezers, chillers, DG sets and power feeds, across one site or fifty. EREMOS V2 builds each alarm from the room's own signals, holds it for as long as you specify before it raises, escalates it to whoever is actually on call, and delivers it by email or WhatsApp. A missing or stale reading is reported as unknown — never as normal. Operating across India and the Middle East.
Live integrations: Modbus TCP · Modbus RTU · OPC UA Client · TCP Listener for devices that push to the gateway — plus mDAQ where a controller exposes nothing useful. Siemens S7, EtherNet/IP and Mitsubishi MELSEC (read-only) are available for the rest of the PLC fleet.
Most sites already have alarms. Hardly anyone trusts them.
A cold-chain site is not short of sensors. Every room has a probe, most compressors have a run signal, the DG set has a controller and the incoming feed has a meter. What the site is short of is a reason to believe any of it at three in the morning. Probes drift. Gateways drop off the network. A sensor battery dies and the last value it ever sent sits in the system looking perfectly healthy. The people on call learn, quickly and correctly, that an alarm is a suggestion.
Then the noise arrives. A door held open for a pallet transfer trips a door alarm every single time, so the door alarm gets muted. A compressor cycling normally trips a threshold, so the threshold gets widened until it never trips at all. Defrost cycles raise temperature alarms on schedule. Within a few months the alarm that matters — the compressor that stopped while the room still needed cooling — arrives in the same channel, with the same tone, as forty alarms that did not matter. It gets the attention it has earned, which is none.
The failure underneath both problems is the same, and it is not a sensor problem. It is that most systems have two states for a condition, true and false, and no way to say I do not currently know. So a stale reading gets treated as a current one, a silent sensor reads as a quiet room, and an old temperature gets combined with a live compressor state to produce a confident answer about a situation nobody is actually observing. A system that cannot say "unknown" will always, eventually, say something false.
"A system that cannot say 'unknown' will always, eventually, say something false."
Three states, not two. And a hold time, so normal operations are not incidents.
A stale reading is never a healthy room.
The outcomes a facilities team reaches for.
The alarm has to survive the network, not just the sensor.
Elpis is deployed across industrial and process operations — mixed-vendor controllers, rotating equipment that cannot fail quietly, and sites with intermittent connectivity. Operating across India and the Middle East. The equipment-alarm capability described on this page runs on that same EREMOS V2 platform. It is built to work offline-first: the EdgeConnect licence validates locally with no phone-home, and per-route store-and-forward preserves every reading through a network or broker drop, queuing locally and replaying in source order on reconnect — so an outage costs latency, not the record of what the room was doing. Every configuration change to EdgeConnect is captured in a hash-chained, tamper-evident change record.
Full operational trust posture → /security · Anonymized deployment patterns → /customers
What facilities teams ask first.
What can you actually read from our refrigeration?
Refrigeration controllers, energy meters and DG controllers over Modbus TCP or Modbus RTU; anything exposing OPC UA over OPC UA Client; and devices that push their own data to the gateway over TCP. Siemens S7, EtherNet/IP and Mitsubishi MELSEC (read-only) cover the rest of the PLC fleet. Where a room exposes nothing useful, mDAQ acquires the probe directly. Bring the equipment list — make and model per room — to the scoping call and we confirm the collection path for each one rather than guessing from a brochure.
What happens when a sensor dies or a gateway drops off?
The condition goes to unknown, and the page and the alarm both say so in those words. It never reports normal in place of a reading it does not have, and it never combines a stale value with a current one. Separately, the signal that went quiet raises its own monitoring unavailable alarm once it passes the window you set for it — so somebody is told that a room stopped being watched, which is the failure mode that actually loses product.
Will this wake people up for things that do not matter?
That is the problem it is designed around. Conditions are compound rather than single thresholds, so a compressor idling on an already-cold room is not an alarm. Every condition carries a hold time, so a door open for a pallet transfer or a scheduled defrost does not raise anything. Maintenance windows suppress routine work while still letting a defined critical set through. You can also see exactly what a rule would do against today's live signals before you save it.
Who gets the alarm at 2am, and how do we know they got it?
Each room follows an escalation chain to an on-call roster with handover, so the alarm reaches the person on duty at that moment rather than a shared group. Delivery is by email or WhatsApp and is tracked, so whether a message was actually delivered is a question with an answer. If the first person does not respond, the chain carries it on.
Our sites have unreliable internet. Does this still work?
Yes — that case is designed for rather than tolerated. Collection and buffering happen at the edge: per-route store-and-forward queues readings locally and replays them in source order when the link returns, and the licence validates locally with no phone-home, so a gateway keeps collecting through an outage. Route buffers are sized by how long an outage they should ride out, and each route reports the coverage it actually has rather than claiming a figure.
Does this replace our BMS or our refrigeration controllers?
No. Elpis sits beside them. The controllers keep controlling — Elpis reads them, and changes no control logic. What it adds is the layer most sites do not have: one vocabulary across mixed equipment, alarms with compound conditions and an honest unknown state, escalation to a real on-call roster, and a history you can go back through after an incident. → /architecture
Looking for the same thing from another angle?
Bring us the room that worries you most.
A room list with what each one has on it, the alarms your team currently ignores and why, and who is on call — that is enough to scope a proof of value. We run it on your real signals.