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.
01 / The platform
Six tightly-coupled modules turn a stream of telemetry into prevented failures and lower bills.
A lightweight agent reads CPU, RAM, disk, network and power — IPMI, Intel RAPL, or a calibrated estimate — and streams it every 30 seconds.
Thermal spikes, zombie processes, idle overconsumption and power spikes are flagged synchronously on every ingest.
A 60-minute-ahead trend forecast with a confidence band, so you stage capacity before the peak lands.
Per-component ETA for thermal and memory drift, with isolation-forest outlier detection fitted per server.
Setpoint, idle-suspend and throttle changes pass rate-limit and guardrail checks, snapshot state, then apply — fully reversible.
When a problem needs hands — a battery swap, fan inspection, fuel refill — Preditoring generates a field-tech work order, idempotently.
02 / The loop
Agents and collectors push server, UPS, HVAC, generator and PDU readings to the ingest API.
Anomaly checks and forecasts run on every reading; predictors estimate failure ETAs.
The suggester proposes actions; policy guardrails approve, block, or route to a human.
Changes apply with a state snapshot, so any action can be rolled back cleanly.
03 / Write-side
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.
04 / Coverage
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.
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
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.
IPMI, Intel RAPL, smart PDU outlet, Modbus, BACnet, SNMP
Reads true draw every 30 s; falls back to a calibrated estimate
Anomaly checks and forecasts run on the same request
Policy engine approves, blocks, or routes to a human
Snapshot, apply, and keep the rollback for seven days
06 / The math
Monthly energy spend before and after the closed-loop optimizer, at €0.18/kWh.
07 / Questions
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.
Spin up a tenant, point an agent at one rack, and watch the savings model populate in minutes.