PreditoringMONITORING · IA · PREDICTION
PlatformPricingSecurityCompany
ENFR
Log inBook a demo
Live · 9 sites · 412 nodes

Know exactly where your power goes — and cut the waste automatically.

Preditoring watches every server, UPS, cooling unit and power strip in real time. It catches energy waste and looming failures the moment they appear, then acts on them for you.

Book a demo →See the platform
Agent deploys in under 5 minutes10–20% typical savingsMulti-tenant & on-prem
ooredoo.preditoring.com2.41 MW
Today cost
€4,820
Savings
14.2%
Open alerts
3
00:0012:00now
rack-B12 inlet 31°C — throttle suggested2m
web-07 idle overdraw · 142 W wasted9m
HVAC-3 setpoint +1.5°C applied22m
10–20%
Energy savings, closed-loop
<30s
Telemetry interval, per node
60min
Ahead power forecast horizon
4-tier
Site → building → room → rack
In production atOoredooOrange TunisieTunisie TelecomSFR Datacenters

01 / The platform

From raw watts to the right action

Six tightly-coupled modules turn a stream of telemetry into prevented failures and lower bills.

01

Telemetry ingestion

A lightweight agent reads CPU, RAM, disk, network and power — IPMI, Intel RAPL, or a calibrated estimate — and streams it every 30 seconds.

02

Anomaly detection

Thermal spikes, zombie processes, idle overconsumption and power spikes are flagged synchronously on every ingest.

03

Power forecasting

A 60-minute-ahead trend forecast with a confidence band, so you stage capacity before the peak lands.

04

Predictive maintenance

Per-component ETA for thermal and memory drift, with isolation-forest outlier detection fitted per server.

05

Policy-gated actions

Setpoint, idle-suspend and throttle changes pass rate-limit and guardrail checks, snapshot state, then apply — fully reversible.

06

Work orders

When a problem needs hands — a battery swap, fan inspection, fuel refill — Preditoring generates a field-tech work order, idempotently.

02 / The loop

Four steps, closed automatically

01Collect

Agents and collectors push server, UPS, HVAC, generator and PDU readings to the ingest API.

02Detect

Anomaly checks and forecasts run on every reading; predictors estimate failure ETAs.

03Decide

The suggester proposes actions; policy guardrails approve, block, or route to a human.

04Act & revert

Changes apply with a state snapshot, so any action can be rolled back cleanly.

03 / Write-side

Every change is gated, logged, and reversible

Before anything touches your infrastructure, the policy engine checks rate limits, single-active-per-target, time-of-day windows and setpoint guardrails. State is snapshotted first, so any action can be rolled back cleanly.

set_cooling_setpointsuspend_idlethrottle_workloadrestart_service
policy-engine · act-4471pending
throttle_workload → rack-B12 · −15% · 40 min
requested
policy_check .......... passed
approval .............. awaiting operator
apply
revert ................ available 7 days
✓Rate limit · 2 of 10 per hour
✓Setpoint within guardrail · 18–27°C
✓Rollback snapshot stored

04 / Coverage

Every box in the room, one pane of glass

Driver-based collectors speak to the equipment you already own — APC, Raritan, SNMP and more. No hardware? Mock drivers run a full demo. Multi-tenant by design, with database-per-tenant isolation.

Servers
UPS
HVAC
Generators
PDU
GPU

Per-outlet watts override estimated server power with measured PDU values.

Role-based access — admin, operator, viewer — re-checked on every write.

Full audit trail on every state-changing request, denormalized to survive deletes.

Database-per-tenant isolation, provisioned automatically on signup.

05 / The measurement chain

A watt is measured, not inferred

Where an instrument exists, Preditoring reads it. The estimate is a fallback, and every figure tells you which of the two you are looking at.

01Instrument

IPMI, Intel RAPL, smart PDU outlet, Modbus, BACnet, SNMP

02Agent

Reads true draw every 30 s; falls back to a calibrated estimate

03Ingest

Anomaly checks and forecasts run on the same request

04Decide

Policy engine approves, blocks, or routes to a human

05Act

Snapshot, apply, and keep the rollback for seven days

Where this differs from ordinary monitoring
Generic monitoringPreditoring
PowerGeneric monitoringCPU utilisation used as a proxyPreditoringIPMI or RAPL per host, overridden by measured PDU outlet watts
FailureGeneric monitoringAn alert once something has already brokenPreditoringA per-component ETA, with the work order written before it does
ResponseGeneric monitoringA page to whoever is on callPreditoringA policy-gated action that snapshots state and can be reverted
CostGeneric monitoringNot modelled at allPreditoringPer-server cost at your own tariff, and the bill avoided
ScopeGeneric monitoringServers, because that is where the agent runsPreditoringUPS, HVAC, generators and PDUs on the same pane

06 / The math

A typical 1.5 MW facility

Monthly energy spend before and after the closed-loop optimizer, at €0.18/kWh.

Before Preditoring€148,000
With Preditoring€127,000
€252ksaved per year · 14.2% lower
Where the savings come from
Idle reclaim€7.4k35%
Cooling optimization€6.1k29%
Peak shaving€4.8k23%
Failure avoidance€2.7k13%

07 / Questions

The things ops teams ask first

The agent reads true draw via IPMI, falls back to Intel RAPL, then to a calibrated linear CPU-load estimate. Where smart PDUs are present, measured per-outlet watts override the estimate entirely.

See your own load in Preditoring

Spin up a tenant, point an agent at one rack, and watch the savings model populate in minutes.

Book a demo →See pricing
PreditoringMONITORING · IA · PREDICTION

Intelligent infrastructure power optimizer. Multi-tenant, on-prem ready.

Platform
How it worksPricingSecurity
Company
AboutContact salesBook a demo
Legal
PrivacyTermsDPA
© 2026 · preditoring.comContact