Metis

One operating objective across the site.

Metis turns site telemetry, controls, tariffs, contracts, and operating history into a digital system model. Foundation models and agents explore possible futures; optimization returns the highest-value feasible action for your objective.

Benchmark a siteSee an operating day ↓

The whole site, modeled as it actually operates.

Metis aligns interval telemetry, EMS and controller state, tariffs, market positions, operator actions, outages, and asset limits into one time-consistent operating record. The topology defines the resources, relationships, and POI the program may act on.

SOLARavailable · curtailed · forecastFACILITYfirm · flexible · critical loadGENERATIONfuel · ramp · minimum run · emissionsSTORAGESOC · power · efficiency · wearINTERCONNECTIONimport · export · tariff · marketSHARED BUSphysical + commercial coupling
Illustrative site topology · resources resolved inside one registered POI and control boundaryScroll the drawing sideways to see the full site

Turn the operating record into a decision system.

Foundation models forecast load, generation, price, weather, and outage risk. Agents search candidate programs across objectives and changing regimes. Deterministic optimization rejects anything outside the physical, commercial, or authority boundary.

01

Site operating record

Ingest the operating context the site already produces.

  • EMS + site controllers
  • Interval telemetry + operating history
  • Tariffs + market positions
  • Operator authority + escalation
02

Digital system model

Represent asset state, coupling, contracts, and the POI as one bounded system.

Grid
POI, import, export
Solar
Available generation
BESS
SOC, power, wear
Generator
Fuel, ramp, runtime
Flexible load
Demand, timing, priority
03

Fitted decision program

Use models and agents to search possible futures and strategies inside the site model.

  1. 01Foundation-model forecasts for load, generation, price, weather, and outages
  2. 02Scenario generation across uncertainty and regime shifts
  3. 03Agent search across candidate strategies and objectives
  4. 04Deterministic feasibility and permission checks
04

Best feasible action

Return the highest-scoring action that clears the full boundary, with evidence.

  • Buy or export
  • Charge or discharge
  • Generate or hold
  • Shift, curtail, or preserve
  • Escalate or abstain

Evidence gate + guarded refit

Every candidate is replayed before promotion. Expected and realized outcomes can return a program to shadow, operator approval, or refitting.

Site data → digital system model → fitted decision program → bounded operating decision

Prove the program before it earns authority.

Replay candidate decisions against the same intervals, constraints, tariffs, and baseline. Run live in shadow. Promote only actions that clear agreed performance and control thresholds.

  1. 01

    Replay

    Run the candidate against the same intervals, constraints, tariffs, and baseline decisions.

  2. 02

    Shadow

    Publish timestamped recommendations beside current operation and score the realized outcomes. No control.

  3. 03

    Bounded execution

    Promote only actions that survive agreed thresholds, permissions, abstention, rollback, and monitoring.

Optimize the horizon. Advance one action.

The fitted program resolves cost, market value, resilience, emissions, fuel, and asset life across the full horizon. It advances the next bounded action, ingests the outcome, and solves again as site state changes.

Illustrative 24-hour operating day

Four assets stacked under one load curve, read against one firm import limit. Hover the chart — or focus it and use the arrow keys — to read any hour.

Illustrative microgrid operating day: grid import, generator, solar and storage stacked under the site load curveOvernight the site is served entirely from the grid at about 3 megawatts. Solar carries the middle of the day and charges the battery with 11 megawatt-hours. From 16:00 to 22:00 net grid import sits exactly on the 4.0 megawatt import limit while storage and the standby generator supply the rest of the evening peak, which reaches 7.2 megawatts.

Hour

18:00

binding window

Site load

7.2 MW

 

Solar

0.4 MW

 

Storage

2.0 MW

discharging

Generator

0.8 MW

 

Net grid import

4.0 MW

at the limit

Illustrative schedule · not measured customer performance · 5 MW solar, 2.5 MW / 10 MWh storage, 1 MW standby generator, 4.0 MW firm import limit · hourly averages, energy balance closed to 0.1 MW

One fitted decision, with every binding constraint attached.

The returned schedule carries the limits that bound it, the headroom it preserves, the permissions it requires, and the reason to abstain or escalate.

Constraint-coupled site schedule

One feasible next-interval plan, resolved against the site's current physical and commercial state.

Illustrative operating state

Next interval

16:15–16:30 local

Feasible · 2 binding
Grid import
3.15 MW
Solar accepted
2.20 MW
BESS discharge
0.55 MW
Generator
0.80 MW
Load served
6.70 MW

Onsite supply and import sum to the load served

Timestamped recommendation routed to the existing operating workflow.

Approval required

POI & interconnect

Import ≤ 4.0 MW

Export ≤ 2.5 MW · hard limit

0.85 MW to the limit

Tariff & demand charge

Binding

3.2 MW demand target

16:00–21:00 peak window

0.05 MW to the target

Resilience reserve

Binding

SOC ≥ 68% at 17:00

Islanding reserve · firm

68.4% projected

Battery state & wear

72% SOC now

1.8 MW discharge headroom

1.25 MW unused

Generator fuel & ramp

90 min minimum run

0.3 MW/min ramp · fuel curve

52 min still committed

Solar & load forecast

2.2 / 6.7 MW

P50 next-interval forecast

Solar fully accepted

One interval, six bounds · accent marks the two the plan is up againstIllustrative · no control path shown

Put one operating decision into replay.

Register the baseline, connect the required data, and compare a fitted program with current operation before anything touches control.

Benchmark a siteSee deployment →
Peak and curtailment on the same site →