Networks and data centres, one governed platform
AEON monitors branch and enterprise networks today, and brings power, cooling and IT together for data centres. Everything runs on your own infrastructure.
Built for large, spread-out networks
Banks, PSUs, utilities and enterprises with hundreds of sites, and teams who need to know which branch is down before the branch calls.
Health at every level
Organise devices by head office, zone, cluster and branch, or any structure you use. See link and device health rolled up at any level, with problems first.
Correct history through change
When a branch moves between zones, reports keep its past with the old zone and its future with the new one. No rewriting history.
Traffic analysis
NetFlow v5 and v9, IPFIX and sFlow from routers and firewalls: top talkers, services, conversations and the busiest interfaces per location.
Discovery and topology
Find devices from seeds and address ranges, follow LLDP and CDP neighbours, review what was found, then approve it into inventory with a topology map.
One alert per fault
When a zone hub fails, every branch behind it goes quiet too. AEON sends one alert naming the hub and holds the branch alerts back, kept and visible, with the cause stated. Maintenance windows suppress without losing the record.
Runs in pairs, upgrades in place
Run two of each service for redundancy: the worker takes one organisation at a time, so instances share the estate instead of each sending every notification. Upgrading replaces them one by one — a build refuses to start against a schema it cannot use, and drains before it exits so requests in flight finish.
Backups you have actually restored
One command backs up and then restores the copy into a scratch database to check it: row counts against the manifest, the evidence chain re-verified from the restored rows, and the role AEON runs as confirmed able to read the result. A backup nobody has restored is a hope, not a backup.
SLA reports by business hours
Availability inside and outside your working hours, holidays and time zones included, with downtime, outages and time to restore. PDF, Excel or CSV, on a schedule.
Services, as a branch sees them
Web, port and DNS checks run from wherever the collector sits, so you learn whether internet banking works from a branch, not just from the data centre. Certificate expiry is warned about weeks ahead.
Wireless
Controllers, access points and each radio band separately, because a 2.4 GHz radio can be saturated while the access point looks healthy. Cisco and Aruba controllers are supported.
Your service desk, kept current
ServiceNow and BMC Helix: one incident per fault, updated while it continues and closed when it clears, with AEON's reasoning attached. Credentials never sit in the database.
IT, power and cooling in one view
Monitoring, governed control, AI diagnosis and governed execution are built and tested. Servers are read and acted on over Redfish and IPMI; UPS, PDUs, meters and cooling are read over Modbus; building management systems are read over BACnet. Every protocol is tested against a simulator rather than the equipment itself, so validation against physical hardware is the next step and we say so rather than imply otherwise.
Unified monitoring
Servers over Redfish and IPMI, facility plant over Modbus, network over SNMP, plus syslog and SNMP traps. Power, temperature and battery readings sit alongside network and IT health in one model.
Governed control
Every command passes a role and risk matrix and a physics envelope. Automation is capped below a human administrator, and irreversible actions need two approvers.
Tamper-evident record
Each request, decision, approval and refusal is written to a hash-chained ledger you can verify at any time.
Predictive intelligence New
Conservative anomaly detection, failure forecasting and topology-aware root-cause correlation, with the reasoning shown.
Self-healing
Remedy requests conservative fixes through the same governance gates as a person, within a scope an administrator approves. Approved actions are carried out by an executor that leases the credential once, verifies the outcome, and escalates rather than retries when a device refuses.
Hardware and facility integrations
Redfish for current servers, reading health and performing governed power and boot actions. IPMI for the servers that predate it, authenticated and encrypted, read-only. Modbus for UPS systems, PDUs, energy meters and cooling, driven by register maps so a new model is a file rather than code. VMware through vCenter for hosts, guests and how full each datastore is. BACnet for building management systems: AEON finds the controllers, asks each what it contains, and reads it. It has no way to write over BACnet at all, so it sits alongside a building system's certified controls rather than competing with them.
Governed autonomy, end to end
Automation does the tireless work; people stay in command of anything that matters.
Discover and organise
Find devices, approve them into inventory, and place them in your own organisation structure.
Monitor
Collectors at each site poll devices and receive flows, buffering locally so an outage at the centre loses nothing.
Alert and report
Rules evaluate continuously, notifications go to the right people, and SLA reports arrive on schedule.
Act under governance
Control actions are checked on role, risk and physics, approved by people where required, and recorded.
See it on your own estate
Start with a monitor-only pilot and watch AEON work against your real network or facility.
Request a pilotSelf-healing, with a leash
Most faults in a data centre are dull and repetitive — a fan deviating, a service that stopped, a threshold drifting. AEON's Remedy module requests fixes for those itself, and an executor carries out what governance approves. The interesting part is not that it acts; it is everything it is prevented from doing.
Diagnose before acting
Cortex correlates symptoms across the fleet before anything is attempted. A rack running warm because a CRAH failed is a different fault from a rack running warm under load, and the remedies are not interchangeable.
Bounded by design
Remedy holds a capability scoped to a verb, a device and a reversibility ceiling, expiring in minutes. It cannot escalate beyond that ceiling even if something convinces it to try.
Some things it may never do
Breaker operations, fire suppression, generator stops and UPS bypass are withheld from automation entirely. Those need an operator at the panel, and no configuration change makes them available.
Physics has the final say
Every remedy is checked against the plant's thermal and electrical envelope. A fix that would resolve the fault and breach the envelope is refused, and the fault escalates to a human instead.
Irreversible needs two people
Anything that cannot be undone — a firmware push, a topology change — requires two independent approvers. Service-affecting actions require one.
Everything is written down
Each action, each refusal and each approval is hash-chained. The refusals matter most: they are how you demonstrate the control was live and working.
Cost, not just efficiency
PUE is a ratio, and ratios don't appear on an invoice. What appears is kilowatt-hours multiplied by the tariff in force when you drew them, plus a demand charge set by your worst fifteen minutes of the month. AEON works that seam from measured power: your tariff, your readings, your bill. Acting on what it finds is a separate decision that goes through the same two gates as any other command.
Your tariff, not an average
Time-of-day rates, the demand charge, fixed charges and taxes. Energy is priced at the rate in force at each reading, in your local time, so peak and off-peak are separated rather than blended. A tariff is refused unless its periods price every hour of the week: a gap would produce a bill that looks plausible and is wrong.
The fifteen minutes that set the charge
Crossing your billed maximum demand raises the charge for the whole period, not just the block that caused it. AEON reports the billed peak, when it happened, and how far above the next separate window it was — the gap being the part worth attacking, priced in your currency.
Warned while you can still act
Inside the block in progress, AEON projects where the average will land if load carries on as it is, compares that with the peak already set this period, and tells you the ceiling to hold for the remaining minutes to stay under it. Reporting afterwards that the charge was set is useless. This reads meters and touches nothing, so it needs no approval.
PUE, or an honest refusal
Total facility power divided by IT power, where the meters exist to compute it. Where either measurement is missing you get the reason rather than a number, because a PUE derived from a guess is worse than no PUE.
Opportunities that state their assumptions
Each finding says what it is worth and what it takes for granted. The cost of moving a kilowatt-hour out of the dear period is arithmetic; which of your load is actually deferrable is not something AEON knows yet, and it says so instead of assuming.
Safety always wins
Every energy proposal passes the same two gates as any other command. A setpoint that saves money but breaches the thermal envelope is refused on physics. Cost optimisation never overrides the hall.
Built by people who sell power
Aevora's founding team develops and operates generation assets. Tariff structures, open access and demand charges are not a feature we researched — they are the business we came from.
Pre-cooling — cooling a hall deeper while energy is cheap so it coasts through an expensive slot — is costed, physics-checked, and can now be carried out: AEON writes the setpoint, reads it back to confirm the plant took it, and records what to restore and when. It will only write a point an administrator has explicitly permitted, within bounds recorded alongside it, and no setpoint has yet been written to real cooling plant — validate on one controller first.
Ask before you act
Can I take chiller two down on Thursday? What breaks if I lose feeder B at this load? Will another forty-kilowatt rack fit in row D? Most tools answer from the nameplate ratings on a drawing. AEON answers from what the building is actually doing: the measured draw on each path, the capacities you have recorded, and an explicit list of what it could not determine.
Maintenance windows
Take a cooling unit or a feeder out on paper first. AEON measures the load it actually carries, checks whether the equipment that would take it has the recorded capacity, and answers yes, yes with the risk named, no, or — where a capacity has never been recorded — unknown, naming what to fill in.
Failure rehearsal
Lose a feeder or lose a cooling unit and see where redundancy actually runs out rather than where the single-line diagram says it does: which devices have a second source, which do not, how much load transfers, and how much simply goes dark. Compound failures — losing two things at once — are in development.
Capacity that isn't stranded
Racks are routinely provisioned against nameplate draw rather than measured draw, which leaves real capacity unused. AEON answers from measured draw and tells you which constraint binds first: the site's design load, the cooling that has to remove the heat, or the rack units actually free.
An answer is never a permission
These questions are answered by reading; nothing here touches plant. A scenario that looks fine is a projection from measurement, not an authorisation, and the real command still goes through both gates. Where the margin after a transfer is thin, or intake temperatures are already high, the answer says so rather than leaving you to notice.
A platform that tells you when it is wrong
Every monitoring vendor claims accuracy. None publishes it. AEON records each prediction and each root-cause call with the evidence behind it, then scores it against what actually happened — and shows you the misses.
Its own error rate, in the product
"This will reach 35 °C in eighteen minutes" is checked against whether it did. "That cooling unit is the cause of these 250 alerts" is checked against whether they cleared when it did. One endpoint returns the rate, broken down by claim type, with the recent wrong calls and why they were wrong.
Silence is never counted as success
A prediction the data cannot settle is reported as unresolved in its own category, never as correct. The tolerances are fixed in the code rather than in a setting an operator could tune until the score improved.
Why it matters to you
The first question after any demo is how often it is wrong. With AEON that stops being a sales answer and becomes a number you check yourself, on your own estate, after a month.
Servers need no vendor integration at all
Every enterprise server of the last fifteen years has a management controller that speaks Redfish or IPMI. AEON reads those directly, so Dell, HPE, IBM, Lenovo, Oracle and Cisco UCS in the same hall are one collector and one credential policy — not six integrations.
Servers
iDRAC, iLO, XCC, ILOM and CIMC, over Redfish where the firmware offers it and IPMI where it does not — so a mixed estate of ages works without configuration. Power state, inlet and exhaust temperature, fans, voltages, power draw, disk and memory health, and the event log. Read-only unless an administrator deliberately enables more.
Network
Cisco IOS and NX-OS, Juniper Junos, Aruba and Cisco wireless, with profiles for the vendor detail. Everything else is read through the standard MIBs, which cover interfaces, errors, discards and hardware sensors on almost any switch — including equipment whose manufacturer no longer exists.
Power and cooling
UPS through the IETF standard MIB, so Eaton, Riello, Socomec and Vertiv are one profile, with Schneider's own MIB alongside it. Chillers, room units, PDUs, energy meters, generating sets and battery plant over Modbus; building systems over BACnet, read-only beside their own controls.
Rack PDUs and environmental sensors vary by model and firmware more than anything else in a data centre, so AEON captures them on site rather than shipping a profile written from a document. It takes about half an hour per device family, and the step that cannot be skipped is comparing a live reading against the unit's own display — a profile that reads deciwatts as watts produces a dashboard that looks right and a capacity decision out by a factor of ten.
Whether bandwidth is actually the problem
A utilisation alert says a link crossed 80%. That is a number, not a diagnosis. AEON reads the interface counters and the flow record over the same window and says which of three different problems you have.
Saturated, or only briefly full?
Discards while a link averages 10% mean it is not full — it is full for milliseconds at a time, and a 1 G to 10 G upgrade often changes nothing. The cause is usually a backup, a replication window or a storage sync bursting at line rate, and the fix is at the source.
Or is it a cable?
Errors rising on a quiet link are physical: a duplex mismatch, a dirty fibre, a failing optic. AEON says so, and tells you that errors on one side only usually mean the cable while errors on both usually mean the link settings. No amount of capacity planning fixes a bad patch lead.
And it changes nothing
AEON alters no routing, no QoS and no switch configuration, by decision rather than by omission. Automated traffic remediation on a production network is how an automation system turns a slow afternoon into an outage. It names the link, the duration and the talkers; your engineer makes a change they understand.
Evidence your regulator verifies without touching your systems
A review normally means giving an auditor a login and letting them take screenshots. AEON exports a period as one signed artefact instead — and they check it on a laptop.
One signed record of the period
The governance ledger with its hash chain, every approval and every refusal, availability from the alert record, the platform's own self-check and its accuracy. No credentials, no telemetry, no device addresses: it is made to leave the building.
Checked without installing anything
The verifier is one file of standard Python that imports nothing from AEON. It recomputes every digest and the ledger chain from the entries themselves, and reports removal, alteration or reordering. An auditor should not have to install the system being audited.
A gap is reported, not smoothed
If the record itself has a break, the bundle says so, still exports, and signs that admission. Concealing it would make the whole exercise worthless.
One set of meter readings, three answers
What it cost, what it emitted, and how much of it your own generation covered — from the same measurements, so the three numbers agree.
Carbon by the hour, not by the year
Grid intensity applied at the hour the power was drawn. In testing, the same energy drawn at night and at the evening peak differed by more than a factor of two — a difference an annual average cannot see. Renewable certificates are reported beside the measurement, never netted against it.
Your own supply, matched to your load
Declare captive generation, open-access blocks and the grid. AEON shows how much of your load your own supply covered, and how much generation went unused while you were buying from the grid at another hour. Cheapest, cleanest and best-covered are reported as three separate answers, because they are rarely the same block.
Built by people who run generation
Aevora's founding team develops and operates hydropower. The question this answers — why are we buying power at seven in the evening when our own plant was spilling at noon? — is one we had before we built it. It neither dispatches generation nor trades: a control room operates under a grid code, and a monitoring platform has no business issuing set points into it.
It learns your building from your building
Most thermal models come from drawings and start drifting the day they are made. Every cooling reduction your hall has already lived through is an experiment that has already been run.
Constants measured from history
How fast the hall warms when cooling is reduced, and how fast it recovers — measured from works windows, trips and setpoint changes already in the record. Nothing is published from fewer than three episodes, the disagreement between them travels with the figure, and it goes stale after six months because buildings get re-arranged.
An engineer always outranks the measurement
Where someone has recorded a value by hand, that is what is used. Where the measurement disagrees materially, AEON says so: one of the two is wrong about this building, and that sentence is usually worth more than either number.
Say what should be true. AEON keeps checking.
A threshold tells you when a number crosses a line. An intent describes a condition that is supposed to hold — and that decays quietly when nothing breaks.
The drift nothing alarms on
A feeder re-patched during a works window and never patched back. A hall filling one server at a time. Nothing is down, nothing pages, and the estate is one failure from an outage. Declare that every server keeps two sources, or that no site runs above 80% of design load, and AEON reports when it stops being true.
Unknown is a real answer
An intent that cannot be checked — no readings, no recorded design load — reads unknown and never as holding. Silence must not read as success, or the feature becomes a way of feeling covered rather than being covered.
Nothing acts without the gate
Where AEON has a proposal it says so, and says it goes through the same approvals, physics checks and ledger entry as any other command. Intent without a gate is automation with better marketing.