Products · Software
Kinetra™ — Operations for the AI Factory
Compute, cooling and power held as one connected model of the facility, on one clock. AI predicts and prices what is coming. Everything reads the model — one governed path writes.
Request a Demo →
Kinetra™ — the facility as one connected model
See · Decide · Act
Why We Built Kinetra
From Power Systems to the Whole Floor
An opportunity to disrupt and optimize data center operations
Compute, cooling and power are still run by separate teams on separate tools. We already build control software for grid-scale electrical equipment, and saw the chance to bring those layers together into one operation.
End-to-end traceability through a knowledge graph
Every rack, GPU, cooling loop, breaker, job and tenant is connected in one graph. Any event traces from the equipment to the workload and customer it affects, and back again.
The Platform
One Platform. Three Planes.
An AI data center is three machines pretending to be one — workload, facility and power, watched by three teams at timescales a thousand times apart. Kinetra holds all three as one graph. Each plane stands on its own; the first one is a complete product by itself.
See
The facility as one connected knowledge graph — racks, GPUs, fabric, cooling, distribution, breakers, jobs and tenants on a single clock. Plus the work that sits on it: alerting, tickets, runbooks, tenant billing.
Decide
Correlates across every layer, predicts power and thermal events, and prices the options. Proposes actions as structured intents — and never actuates.
Act
The only path that writes to anything. Every change passes the Fabric Gate: closed action vocabulary, interlock checks, tiered approval, tamper-evident ledger.
The Fabric Gate
The single checkpoint between any proposal and any equipment. A model cannot be audited, so we took it out of the write path entirely.
For Data Centers
Which Tenant Is Affected, Right Now.
A training job changes phase and the draw swings hard. Kinetra holds the job, the cooling response and the distribution path as one event on one clock — so the answer takes a click, not three people comparing notes afterwards.
Key Features
Why Choose Kinetra?
One Data Model, One Clock
Not one dashboard — one model. Workload, facility and power stored as entities and the relationships between them, so a breaker traces to a tenant.
The Model Never Touches a Breaker
The system proposes. A named human approves. Execution runs through the Fabric Gate. There is no other write path.
Cross-Layer Root Cause
Incidents that cross compute, cooling and power resolve in one place, instead of three teams comparing notes after it is over.
Prediction, Not Reaction
Power and thermal events forecast and priced ahead of time, with the options laid out as structured intents.
Provable Service Levels
Tenant impact, chargeback and compliance evidence assembled from the graph — reports nobody has to build by hand.
Built by Power People
TeraStor Energy already writes control software for grid-scale electrical equipment. The write path is built to that standard.
Integrations
It Connects. It Doesn’t Replace.
Your scheduler, your building systems and your electrical plant stay where they are. Kinetra reads them into one graph — and writes back through one gate.
Your Scheduler Stays
Kinetra connects to the MLOps and scheduling stack you already run rather than replacing it — and tells it which racks have electrical and thermal headroom.
Facility & Building Systems
BMS, CDUs, DCIM and metering read in over the protocols they already speak — Modbus, BACnet, DNP3, REST.
Electrical Plant
Switchgear, busway and distribution modelled with their interlocks, so a proposed change is checked against the real plant before anyone approves it.
Adoption
Start by Watching. Automate When You Decide To.
Nobody grants software authority over an electrical room on day one. Each stage is your explicit decision, tier by tier.
Stage 1
Observe
Kinetra runs read-only. The graph proves itself against your own instruments.
No writes, no risk, often no change board.
Stage 2
Advise
The system recommends. Every action is approved by a named human before it executes.
Trust earned in stage 1.
Stage 3
Automate
Low-consequence tiers run on their own, inside the same gate and the same ledger.
Your explicit decision, tier by tier.
The 30 day pilot
One slice of the fleet, in observe mode, tracing one workload end to end. Your team checks our graph against your own instruments.
Get Started
See Kinetra on Your Floor
Pick one slice of the fleet and we’ll run Kinetra against it in observe mode for thirty days. Design-partner terms are open this cycle.