VoltMap

How it works

Three things work together: a model, a forecast and a team of agents.

VoltMap keeps a working model of your building, uses it to see the next three days, and gives each major piece of equipment an agent that knows what normal looks like.

Digital twin: a working model of your building.

The twin learns how your facility uses power by hour, by day and by type of equipment. It starts from the data your utility already records, so there is something to work with on day one.

Add sensors or a building management system and the twin becomes physics-based: it models how heat moves through rooms and product, which is what makes pre-cooling safe to plan.

What goes in

15-minute meter data
Up to 24 months of bills
Weather history and forecast
Optional: sensors and building systems

The twin

  • Load by equipment type
  • Thermal behavior of rooms and product
  • Recalibrated as new data arrives

What comes out

A 72-hour demand forecast
What-if tests of any schedule
The baseline that savings are measured against
Flags when something departs from normal
15 min
resolution of every prediction
72 hours
forecast horizon, refreshed daily
Every run
the twin reports its own accuracy and data quality

Predictive energy optimization: act before the peak, not after the bill.

A demand charge can't be undone once the peak has happened. VoltMap forecasts demand for the next 72 hours with a confidence range, and when a costly peak looks likely it plans the response in advance and tests it on the twin first.

72-hour demand forecastSample data
100200300kWTodayTomorrowDay 3alert sent 18+ hours ahead
  • Forecast demand
  • Confidence range
  • Alert sent
  • Predicted peak
The same day, two schedulesSample data
peak-price hours10020030000:0006:0012:0018:0024:00
  • Usual schedule
  • Optimized schedule
  1. Forecast

    Demand, weather and prices for the next 72 hours, in 15-minute steps.

  2. Simulate

    Candidate schedules are run on the twin: pre-cool, stagger, shift. Any that would breach a limit are dropped.

  3. Recommend

    You receive the best plan with its reason and expected saving, at least 18 hours before the peak.

  4. Verify

    Afterwards the result is compared with the forecast, and the twin learns from the difference.

Agent runtime: every major piece of equipment gets its own agent.

Each agent learns one machine: its normal power signature, its schedule, how it responds to weather. It diagnoses itself when something changes and reports upward. A site coordinator weighs the reports, settles conflicts between machines and checks every proposal against your limits before anything reaches you.

Self-diagnosis, in practice

An agent compares its machine's draw with what the twin expects for today's weather and workload. When the gap holds for three days or more, it works out the likely cause from the pattern, estimates what the fault is costing, and says what to check.

Refrigeration power, last 30 daysSample data
Normal range23% above normal

Local knowledge

A site-wide average hides a failing compressor. An agent that watches one machine notices.

No silent actions

Agents propose. What they may do on their own is set by the autonomy level you choose, and everything is logged.

Open to other systems

The same findings are available to your maintenance, accounting and automation tools through MCP and APIs.

See the numbers for your own sites.

Send three months of bills from up to three locations. We'll show what was overpaid and why, using your own figures.

Request a demo