Every departure.
Less shared power.
Work out who needs energy, by when, and the smallest connection that can serve them. Compare charging policies, inspect the physical limits, and replan when a charger fails.
The physical limits come first.
Each charger has its own rating and connector. Vans keep their assigned plug until departure. The planner respects battery limits, charging efficiency, taper, outages and the shared site connection.
FCFS, earliest-deadline-first and least-laxity-first use the same constraints. OR-Tools builds the optimized plan; an independent validator checks the allocations.
Same energy targets. A lower peak.
In the four-van example, every van receives its requested energy: 84 battery kWh in total. The three greedy policies peak at 22 kW; the optimized schedule peaks at 8.89 kW.
Four vans · two fixed 11 kW plugs · 90% charging efficiency.
How much shared power is enough?
The afternoon pair needs 48 battery kWh in six hours. With 90% efficiency, that needs 53.33 grid kWh. Below 8.89 kW, there simply isn't enough energy.
Explore the six-hour energy requirement here, then use the app to include charger occupancy, battery taper and the complete roster.
Battery energy available = connection × 6 h × 0.90
Questions become useful actions.
NVIDIA Nemotron on Nebius Token Factory reads computed schedule facts and can request an outage replan through typed tools. The optimizer creates the allocations, and independent checks return the updated result to the dashboard.
In the CH-01 outage example, every departure remains met at a 10 kW peak. The animation follows this same tool loop.
Forecasting for future price windows.
In the separate forecasting study, TimesFM and Chronos used 37 held-out days of historical German prices. Chronos Small reduced price MAE by 29.2% versus ridge; the extra charging-cost benefit was small in the four-van replay.
The study uses 2019 price data. Published day-ahead prices feed known windows directly. Explore the study and results.
Use the editable application.
Open the live recording session · Download the demo narration and actions
This hosted page explains the architecture and workflows. The complete planner runs locally from the downloadable build, with the React dashboard, FastAPI, Pandas and OR-Tools.
- Download and extract the test build.
- Install Python 3.11+ and uv, then run this command inside the extracted folder:
uv run --locked python scripts/run_demo.py
Open http://127.0.0.1:8000. Start the guided demo, edit the inputs, optimize, check capacity and simulate an outage. Node.js is unnecessary for this prebuilt package.
Planning runs without an API key. The assistant can summarize the current results locally or connect to Nemotron through the private backend.
Source, film and voiceover
Downloads and editable animation · Timestamped voiceover text · MIT source repository
The cost example uses synthetic prices, an illustrative €15.20/kW/month tariff and a zero prior peak. Its €355.87 → €159.20 result covers the example's partial billing period.