At a convenience store, a refrigerator usually has one assignment: keep the drinks cold. Maplewell Energy gave it a second job. Before a building’s electricity demand was expected to peak, refrigeration equipment at two Parkland USA sites pre-cooled its contents. During the costly interval, the compressors could ease off while the cans and bottles coasted on stored chill. Nothing resembling a utility battery appeared in the aisle. Yet electricity had effectively moved through time.
Maplewell reported that the 2023 pilot, conducted in Denver and Salt Lake City, cut peak energy use by 25 percent. The company projected that adding battery storage could push the reduction to 42 percent, though that higher figure was a forecast, not a field result. The distinction matters. Maplewell is selling an energy-management system, not magic. Its work depends on tariffs, physics, controllable equipment and the tolerance of the operation underneath it.
That practical constraint is also the company’s appeal. Founded in Colorado in 2019, Maplewell builds JANiiT, a software-and-control platform for distributed energy resources. Batteries, solar inverters, backup generators, HVAC and refrigeration feed it data. JANiiT forecasts what is likely to happen, calculates a dispatch plan and sends commands locally or through the cloud. The objective may be a smaller demand charge, a better time-of-use bill, revenue from a grid program or enough reserve to ride through an outage.
The expensive hour
Electricity is not priced only by how much a business consumes. Many commercial tariffs also punish the highest interval of demand in a billing period. One badly timed surge can shape a month’s bill. Meanwhile, a battery may have several possible jobs: shave a facility peak, buy power when rates are low, preserve backup capacity, respond to a utility event or trade in a wholesale market. These goals can conflict. Discharge too soon and the battery may be empty when the real peak arrives.
JANiiT is built around that scheduling problem. Maplewell’s published specification describes probabilistic forecasts for demand, generation and market prices 12 to 24 hours ahead, with online optimization every five minutes. A local controller can execute load-following decisions in less than 40 milliseconds. The long view predicts the weather of the day; the short loop keeps a physical system inside its guardrails.
The stack is deliberately split into three layers. JANiiT Cloud holds the customer dashboard, database, forecasts and optimization. Cloud Edge brings models and high-fidelity data storage onto an on-site AI computer. The controller and metering layer handles deterministic control and can keep running when a network becomes intermittent. A customer can deploy the components together or use them separately.
One decision, three timescales
Forecasts demand, solar output and prices. Coordinates a portfolio.
Runs models near the data when latency, privacy or bandwidth matters.
Executes deterministic commands and falls back locally if the network fails.
This architecture lets Maplewell sell the same core system into very different conditions. A solar-and-storage developer can simulate project revenue before specifying an EMS. A portfolio owner can coordinate batteries across multiple sites. A facility manager can attack demand charges. A utility can aggregate resources into a virtual power plant. A defense installation can run an air-gapped controller where satellite or radio may provide only one-way monitoring.
From a national lab to the loading dock
Maplewell’s technical lineage runs through the National Renewable Energy Laboratory. In 2022, the company secured an exclusive license to OptGrid, distributed-control technology created under the Department of Energy’s ARPA-E NODES program. OptGrid manages energy from the bottom up: devices solve local control problems, then coordinate rather than waiting for one distant brain. NREL says the approach has been tested in laboratory simulations of more than one million devices.
Emiliano Dall’Anese, Maplewell’s chief technology officer and senior adviser, helped invent the technology. His stated aim is infrastructure that is flexible, reliable and clean through decentralized control. In product terms, the license gives JANiiT a way to bridge individual assets and large aggregations without assuming perfect connectivity. That is useful on a commercial campus. It is essential on a ship or isolated microgrid.
“Without this program, we would have not explored LLMs in our application.”Matthew Irvin, on Intel Liftoff
The company’s modeling path changed again after it joined Intel Liftoff in 2024. At an Intel-hosted program, Maplewell experimented with language-model techniques for time-series forecasting and later reported a 25 percent accuracy improvement over traditional methods. The interesting part is not the fashionable acronym. CEO and co-founder Matthew Irvin said the program made the team comfortable trying something risky that it otherwise would not have explored. Intel also supplied cloud resources, model-optimization tools and edge hardware work.
That is a useful correction to the standard AI story. Better forecasts matter only if they produce a safe command in a real machine. JANiiT supports common industrial protocols including Modbus, MQTT and REST, and published integrations cover solar and battery inverters, backup generators, meters, SCADA and utility management systems. Maplewell’s expertise sits where prediction meets controls engineering.
The proof - and its price
The public cost picture is clearer at the company level than at the customer level. Maplewell opened a $3.5 million seed fund in 2022 with Life Bridge Capital as lead investor. It said the money would support front-end and back-end development, the NREL license, hiring and expansion into industrial and maritime work. Colorado’s Advanced Industries program separately awarded the company a $250,000 grant. Public pricing for JANiiT deployments is not disclosed.
Maplewell makes money through several layers: subscription cloud software, optional controllers and metering, integration and support, and services across the battery-project lifecycle. Its materials list simulation, feasibility studies, financial modeling, procurement, commissioning, ISO integration and predictive maintenance. A 2023 partnership with Urban Electric Power paired JANiiT with rechargeable zinc batteries for turnkey virtual-power-plant projects under an Energy-as-a-Service model. Defense work adds contract-funded research and deployment.
The model is broad because infrastructure is stubbornly specific. Software must know which inverter is installed, what a warranty allows, whether a refrigeration cycle can move, how a utility calculates demand and which cybersecurity rule applies. Every extra integration can slow repeatability. Maplewell’s advantage - meeting assets where they are - is also the operational burden it must master to scale.
Who buys flexibility?
The customer list begins with people who already own or plan distributed energy. Solar and storage developers need to validate a project’s revenue stack. Owners and operators need a portfolio view and reliable dispatch. Energy-service companies need measurable conservation measures. Commercial and industrial facilities want lower bills without interrupted production. Utilities need demand response and non-wires alternatives. Navy users need mission readiness when communications are disrupted.
Maplewell publicly names Parkland USA and U.S. Navy NAVSEA, and its site displays U.S. Navy and NAVSEA marks among enterprise customers. Since 2021, it has developed JANiiT for shipboard and shoreside energy management under Navy small-business research work. The system is intended to optimize storage for shipyards, hybrid electric drives, directed-energy applications and upgrades constrained by existing electrical buses. These are demanding environments where a cloud-only product would be a liability.
In the wider market, Maplewell overlaps with battery-optimization platforms, virtual-power-plant operators and the microgrid suites sold by major industrial-automation companies. Stem’s Athena, AutoGrid’s flexibility software and offerings from Schneider Electric, Siemens and Honeywell represent different versions of the alternative. Conventional building-management systems are another competitor, especially when a customer needs fixed schedules and monitoring rather than predictive dispatch.
Maplewell’s difference is the attempt to keep forecasting, portfolio optimization and deterministic edge control in one system, then carry that system from a convenience store to an air-gapped defense site. The NREL license adds a distributed-control method that is difficult to reproduce from a dashboard alone. The company is not merely observing electricity. It is trying to act before the meter records the mistake.
Where the model fits
- Demand charges or time-varying rates reward better timing.
- Batteries, HVAC or refrigeration offer safe flexibility.
- Sites need local control when connectivity is weak.
- Grid programs pay for reliable response.
Where it weakens
- Flat tariffs offer little value for shifting load.
- Processes cannot move without harming operations.
- Telemetry is poor or equipment cannot accept commands.
- Savings fail to cover integration and battery wear.
The part worth copying
Maplewell’s most portable lesson is not a model architecture. It is the sequence of proof. The refrigerator pilot selected a narrow physical loop, used existing thermal mass, measured a costly peak and changed timing without asking customers to behave differently. Only after that result did the company point toward additional batteries, more sites and virtual-power-plant scale.
A four-step demand-flexibility test
- Find the tariff interval or physical constraint that actually costs money.
- Identify a load that can safely move without changing the customer experience.
- Forecast the event, close the control loop and preserve a local fallback.
- Report achieved results separately from projections that require new hardware.
There is discipline in that last step. A 25 percent field result is more useful than a 42 percent scenario if the reader can tell which is which. Buyers should apply the same discipline to every claim: ask for the tariff, baseline, control window, comfort or production constraints, and the effect of cycling on battery life. A virtual power plant is ultimately a collection of such local bargains.
Maplewell’s wager is that those bargains can become a grid resource. If a refrigerator can pre-cool, a battery can reserve the right hour and a microgrid can keep thinking when the cloud goes dark, new supply is not the only answer to rising demand. Sometimes the overlooked asset is timing - and timing can be programmed.