HyperBase

PLATFORM · ENTROPY™

Agentic orchestration for hyperscale energy.

Entropy™ is a three-layer intelligence system. It observes distributed energy state, models dispatch against physics, and clears setpoints at the tempo of the workload.

01 · ARCHITECTURE

Three layers. One clearing loop.

Hover a layer to inspect it. The system runs top-down on observation and bottom-up on dispatch, with modeling as the closing loop.

LAYER 01OrchestrationBAYESIAN STACKELBERGGAME THEORYHours → Day-Ahead | 1 Strategic AgentCENTRAL BRAINDAMRTMANCILLARYCAPACITYOUTPUTMarket Bids + Portfolio Optim.INPUTBayesian Belief UpdatesMOATGame-Theoretic ≠ Price-TakerDISPATCH SIGNALS ↓↑ GRADIENT UPDATESLAYER 02AggregationPI-MARL +FEDERATED LEARNINGMinutes → Hours | 10-20 Meta-AgentsPHYSICS BOUNDARY — AGENTS CANNOT VIOLATESOLARBESSGASPRIVACYFederated: Gradients Only ↑PHYSICSHard-Embedded in Reward FnCOORDGNN Consensus Across 1000+ DERsSETPOINTS ↓↑ TELEMETRYLAYER 03EdgePINNs ONDISTRIBUTED HARDWAREMilliseconds → Seconds | ~10,000 AgentsPINN-001PINN-002PINN-003PINN-004··· ×10,000 AGENTS ACROSS DISTRIBUTED EDGE NETWORKLATENCY<10ms Control LoopHARDWAREESP32 / NodeMCU ControllersFUNCTIONReal-Time PDE + Anomaly Det.

FIG.02 · THREE-LAYER ARCHITECTURE

HOVER LAYER TO INSPECT

02 · LIVE VIEW

A digital twin of the clearing loop.

High utilization — compute load near peak. Cooling works hard, grid import elevated. Click any component in the scene to inspect its telemetry.

SIMULATED DATA · DEMO

03 · COVERAGE

What Entropy coordinates.

GENERATION

Solar · Wind · Thermal · Gas

Distributed generation assets, dispatched against real-time demand and constrained by local transmission state.

STORAGE

BESS · Thermal

Battery and thermal storage, cycled as the clearing buffer between intermittent supply and hyperscale demand.

LOAD

Grid · Industrial

Dispatchable load across industrial and commercial nodes, bid into the clearing layer as controllable demand.

COMPUTE

Hyperscale · AI

Training and inference workloads, shaped against the available clear to maximize useful FLOPs per MWh.

04 · THE MATH

Physics-informed, multi-agent.

Dispatch policy is learned by agents per asset class, under the constraints of the underlying physics — AC power flow, ramp limits, thermal and state-of-charge boundaries. The policy is optimized against clearing cost, not unconstrained reward.

The underlying approach is documented in the PI-MARL whitepaper: a unified framework for smart-grid markets and hyperscaler virtual power plants.

Entropy™ is licensed per 100 MW of orchestrated capacity.

IP STATUS

Provisional U.S. Patent

“Multi-Layer System for Coordinating Distributed Power Networks.”

DEPLOYMENT

Pilot validation against 1 GW hyperscale asset

Proof of concept modeled against a one-gigawatt hyperscale campus profile.

RESEARCH

PI-MARL · Entropy™ Math

Two published technical documents covering framework and mathematical foundations.