Digital twin
The world builds digital twins of the planet. We build the one that ends at your asset.
Cities, plants and networks already have digital twins: a living replica of the structure that answers an engineering question before the work leaves the drawing board. What that portrait is missing is the layer that changes all of them every day — the atmosphere.
The Atmospheric Digital Twin (ADT) is our answer to that, and it exists: high-resolution numerical modelling, AI and the knowledge of the people who operate, in a model that does not stop at the sky. It ends in your asset registry — at this substation, this cell site, this field, this treatment plant.
And it is not a layer of concept over generic product. In practice, the ADT materialises in Risk360 — the platform that already carries the risk reading down to the asset — fed by our own data engines and read by the specialist AI agents. All of it runs at clients today; the list is at the end of this page.
What it is
The atmosphere is the living layer
A digital twin of a city or a plant is, at heart, a photograph of structure: it changes with construction, retrofit, expansion. The atmosphere changes on its own, hour by hour — rain, wind, heat, smoke — and it is what decides what the structure can take today.
What goes into it
High-resolution numerical modelling, AI, and expert knowledge in continuous learning. The model is ours, and the people operating it correct it: every threshold tuned in the field goes back inside.
It ends at the asset, with a name and a coordinate
A twin of the atmosphere with no ground under it is pretty meteorology. Tied to the registry, it answers about that asset — and that tie is the product. Everything else is input.
Where the world is, and where we are
A digital twin of the Earth is not a new idea, and we will not pretend it is. What is new is the scope: coming down from the planet to the asset. That is where São Tomás is a pioneer, and that is where comparing us to traditional weather stops making sense.
01
The world builds twins of the planet
Destination Earth, the European Union flagship with ECMWF, ESA and EUMETSAT, aims at a digital replica of the Earth system by 2030. NVIDIA built Earth-2. These are public infrastructure and platform efforts, for science and continental-scale forecasting.
02
What none of them does
None comes down to your asset. A replica of the planet does not know where your substation is, what your crew’s safety threshold is, or which ruler your regulator applies. A twin that stops at the sky does not sign a work order.
03
That last mile is what we build
An atmospheric twin coupled to the asset registry, to the operation’s thresholds and to the sector’s regulation — with the AI agent explaining it in the language of whoever decides. Not forecasting in new packaging: a decision about a named asset.
Why now
Risk left the scenario and landed on the balance sheet.
For twenty years climate was a topic for a 2050 report. The last two years billed it early — and over the same period, the ruler measuring companies changed hands.
The climate already billed
The 2023–24 super El Niño
One of the strongest on record. Record heat, severe drought in the Amazon, and a fire season that looked like no previous one.
Rio Grande do Sul, May 2024
The largest flood in the state’s history. Operations, assets and entire cities halted at once — and no risk spreadsheet had that scenario in it.
The 2024 wildfire crisis
Simultaneous fire across several regions, with transmission lines, railways and crops in its path. Those who had the hotspots mapped acted; those who did not, watched.
Over R$ 1 billion a year, in compensation
For breaching outage duration and frequency limits alone (DEC and FEC), Brazilian distributors paid R$ 1.002 billion back to consumers in 2025 — after R$ 1.122 bn in 2024 and R$ 1.080 bn in 2023, per ANEEL. And 2025 was a year of improvement: average outage duration fell 9.2%. It clears a billion even when the indicator improves — and the fine is only the visible part of the bill, before emergency crew mobilisation, corrective pruning and lost revenue.
The ruler changed hands
Restoring service after an extreme event is now a concession target
Decree 12,068/2024 set the guidelines for renewing 20 distribution concessions expiring between 2025 and 2031 — roughly 62% of the market. Among its 17 guidelines is, in those words, an efficiency target for restoring service after extreme weather events. It stopped being good practice and became a condition of the contract.
The financial sector is still bound
BCB 139, CMN 4.943 and SUSEP 666 still require banks, insurers and asset managers to manage and disclose climate risk. If you borrow or insure, the requirement reaches you second-hand.
Adaptation became law
Lei nº 14.904/2024 set the guidelines for climate adaptation plans across all levels of government and across sectors. A plan needs a diagnosis, and a diagnosis needs data per asset.
At the CVM it stopped being a duty and became an edge
Resolução CVM 244, of May 2026, made voluntary the IFRS S1/S2 reporting that CVM 193 would have made mandatory. The pressure did not disappear: it moved. Whoever reports now does it because investors, banks and clients ask — and whoever reports well separates from whoever does not report at all.
And what used to be a project became software
The input got better and cheaper
Machine-learning forecasting and free, frequent satellite imagery collapsed the cost of the raw material. What used to be seven-figure consulting now fits inside software — which is why this sells as a platform, not as a study.
What got cheap is the ground, not the core
Public inputs do not know how it rains in your territory or what your operation’s threshold is. The calibrated model is still ours — what got cheap is the ground it runs on.
The operational copilot stopped being a promise
An agent that reads the risk and explains it in the sector’s language, at operational cost and latency, did not run two years ago. Today it runs in text, in voice and on the field team’s WhatsApp.
Regulation moves. This section reflects what stood in August 2026; if you are reading it much later, check before deciding on it.
Why here
Brazil is the hardest place to solve this. That is why it is the right one.
We are disproportionately exposed and disproportionately underserved. It is both at once that make this the place to build the solution rather than import it.
The exposure is disproportionate
Overhead networks of continental scale, tropical vegetation that grows back all year, and extreme events on the rise. The same storm costs more here than in a country with buried lines and a dormant winter.
Global tools, calibrated for another hemisphere
International climate risk platforms were built for long-term financial risk in northern-hemisphere markets. They carry no Brazilian grid data, they are not calibrated to our rainfall regime, and they do not have ANEEL’s ruler inside them.
The buyer is concentrated
ANEEL ranks 33 large distributors separately — those above 400,000 consumer units. Add transmission operators and the large water utilities and you have a small, nameable universe. That is where we start.
Whoever solves operational climate risk in Brazil solves it for Latin America, Africa and Southeast Asia — the same exposure profile, the same absence of a calibrated solution.
Specialist agents
A specialist AI agent per sector. Not a generic chatbot.
The agents specialise in the sector of whoever is asking — and “specialist” here is not a brochure adjective. It is not the same model with a new name: the vocabulary, the activities, the thresholds and what counts as risk all change. The proof is in the question each one answers.
Power utilities
“Can we do live-line work tomorrow morning?”
Agriculture
“What is the best spraying window this week?”
Water and sanitation
“Can I clear confined-space entry this afternoon?”
Telecom
“Which site loses power first if the front arrives?”
Infrastructure
“Any hotspots near the right-of-way?”
Civil defence
“Which flood scenario does this rainfall trigger?”
Where the team already is
Text
A conversation inside the dashboard, next to the map and the numbers the answer came from.
Voice
The day’s reading spoken aloud, for people whose hands are busy — in the vehicle, at the substation, in the field.
Question and answer in the app the field team already uses. Nothing to install, nobody to train on a new tool.
How much the agent does
Two modes, and the difference between them is how much work you delegate — never how much of the decision.
Copilot
The agent follows along and weighs in: it reads the day’s risk, explains the reasoning and recommends the action. The team still drives — the agent steps in when called, on screen, by voice or on WhatsApp.
Autopilot
The agent runs the whole routine: it monitors, joins the data to the registry, prioritises and leaves the action ready to go. What is left for the person is the one part nobody delegates: deciding.
What the twin is made of
These are the parts, by name. The core is Risk360 — it is there that the atmospheric layer meets the operation and turns into action. The engines are the data products that feed it, and the AI agents are what read it in human words.
Risk360
the core
Where the twin meets the operation: risk per asset, working windows, prioritised action and sustained response during the crisis. The ADT is the atmospheric layer; Risk360 is where it becomes a decision someone signs.
Explore Risk360Weather Trends
forecasting engine
Our own high-resolution forecasting, from minute-by-minute nowcasting to the six-month trend. It is the twin’s pulse — what changes hour by hour over your territory.
Explore Weather TrendsWildfire Sentinel
fire engine
Hotspots, ignition risk and wind-driven spread. It is the layer of the twin that answers for fire, and the one that depends most on current data — fire does not wait for tomorrow’s report.
Explore Wildfire SentinelGeoGuard
hydrology engine
Distributed rainfall–runoff, water level and geo-hydrological risk, with the basin discretised into sub-basins. It is the part of the twin that answers for water: from what falls in the sky to what reaches the intake.
Climate Edge
the long horizon
The same question, decades ahead: climate projection over the real asset portfolio, with uncertainty declared. It is the twin in 2050, for whoever decides where to put what will last thirty years.
Explore Climate EdgeSpecialist AI agents
who reads the twin
One agent per sector, reading the twin and explaining it in the language of whoever decides — in text, in voice and on the field team’s WhatsApp. In MapsAI, the question goes to the map in natural language. A model without a translator does not become a decision.
The ADT is not a laboratory model: it is made of the platforms, the engines and the agents that already run at clients today — the calibrated models, the historical and current data, the agents that explain. The twin is the layer that ties all of it to your asset registry, and it is the layer we keep building from.
Bring us your challenge
If it involves climate, territory and data, it is our ground. The conversation starts from a real slice of your operation — not from a deck.