Ogre gives the energy system foresight, with a dedicated AI engine for every asset, grid node and market. That foresight becomes decisions: constraint actions for grids, schedules for sites, positions for the desk.
−36% forecast error at Enel's DSOs, #1 on accuracy in Elia's blind benchmark. Read the case studies.
Data and telemetry flow in from everything on the grid. A dedicated engine forecasts each one, and the decision layer turns those forecasts into actions: priced, sized, scheduled, and delivered back to the people and systems that execute them.
Volatility became the product. Prosumers bent the load curve, renewables set the price, and last year stopped predicting this year. The methods were never wrong; the scale outgrew them.
Solar, wind, every consumer, every substation, every loss curve. A dedicated AI engine for each, built and retrained automatically, graded daily.
Constraints found and priced, site days planned inside their limits, trades read from distributions. Decisions with numbers attached.
Advise, recommend, signal, or close the loop, from the control room to the trading desk. The audit trail records every step.
Every Ogre AI product terminates in a call someone signs off. Four, straight from the screens:
Gridian prices Thursday 17:00's overload against flexibility and curtailment. The operator approves it on Tuesday, not in the event.
Markets shows the whole distribution before the interval settles: how wrong the forecast could be, and what each side of wrong would cost.
Energy Hub moves flexible load into the solar hours and plans the battery's day, always inside the connection limit.
Portfolio bands turn the peak from a surprise into a known risk with a price: hedged a day early, reviewed against what settled.
Four ways in, depending on where you sit: the portfolio, the control room, the trading desk, the site. Underneath, the same engines, graded daily.
Forecasting as a service for the whole energy value chain: solar, wind, grid load, net load, losses, consumers, EV charging, prosumers and portfolios.
Forecast-led congestion and flexibility management for TSOs and DSOs, from two-week foresight to dispatch signals.
Imbalance every 15 minutes, prosumer generation as a zone product, and country-level fundamentals. All of it as distributions, with the cost of being wrong made visible. Early access now, available from the end of 2026.
Forecast-native optimization inside the connection limit, for data centres, industry and energy communities. Early access now, available from the end of 2026.
Our customers run the value chain. Our partners help us reach it.
Ogre AI integrates its advanced energy forecasting platform with SAP Distributed Energy Resources to bring predictive energy analytics directly into enterprise systems.
Orange and Ogre AI are leveraging AI to optimize energy flows across critical telecommunications infrastructure, including data centres and distributed radio stations, to drive operational sustainability and grid resilience.
Pilots run one to three months, set up in weeks. We'll measure our accuracy against your current method on your own history, and show you the result either way.
About five minutes: your details, your CV, and a few lines on why Ogre. Every application gets a reply.
Forecasting as a service for the whole energy value chain: solar, wind, grid load, losses, consumers, EV charging, prosumers and portfolios. Built, retrained and graded by Ogre, delivered into the systems you already run.
Under-forecast and you buy at balancing prices; over-forecast and you sell at a discount. Every error costs, both ways.
Renewables, prosumers and EVs make every curve weather-driven, and yesterday's profiles stop working.
Tariffs recognize an efficient cost level; losses and imbalances above it stay on your books.
Spreadsheets, static coefficients and engineer-hours: stale within months, wrong on the days that matter.
Technical-loss engines in production for a national TSO and a DSO group, graded against settlement.
National wind forecasting for Poland's transmission system operator.
A large segment of the E.ON PV asset portfolio, expanded after the first engagement.
Consumer and portfolio forecasting for suppliers, one engine per meter where it pays.
One platform, many shapes, and behind each shape a customer: asset owners, retail books, network operators, charge-point networks. This is the actual screen; click a curve or let it tour.
Every asset and consumer gets its own engine: validated against its own history, graded daily, and retrained when conditions change.
Our weather-calibration AI blends six providers, tuned to your geography and equipment: the single biggest driver of accuracy.
One per meter, plant or node if needed. Nobody hand-builds at this scale; the platform does.
Granularities and horizons set per use case, from the next 15 minutes out to long-term planning.
Deliberately careful ranges. The pilot replaces them with your measured number, benchmarked on your own history.
A 100 MW park cutting forecast error by two to three points typically saves this in imbalance costs alone, before curtailment handling and intraday repositioning.
Per-meter engines on a retail portfolio typically cut balancing costs by double digits, and make the prosumer share forecastable at all.
Loss forecasting per network area tightens a DSO's single biggest energy purchase; constraint foresight defers the slow answer: copper.
Forecasts, actuals and accuracy over a clean REST API, for your trading systems, EMS or data platform.
Scheduled deliveries in the formats and cadences your processes expect.
A client-facing interface your teams use daily: forecasts, actuals and accuracy, side by side.
Historical data in, assets registered, engines built: one to two weeks per module, no integration project.
Our forecasts run next to your current method on live data. You see both errors, daily, on your own numbers.
Monthly, yearly or multi-year. If we didn't beat your method, you'll know, and so will we.
We'll benchmark against your current method on your own history, and show you the result either way.
Forecast-led network operations for TSOs and DSOs. Gridian predicts every node of your network, finds tomorrow’s overloads on a physics model of the grid, prices your options, and acts at the level you allow.
Managing the peaks now costs real money, and generators ask why it was them. The bill lands twice: once in constraint payments, once in trust, when curtailment looks arbitrary.
Buying MW ahead of time beats curtailing at the limit, if you know where and when. Procured too broadly it is wasted budget; too narrowly, and the constraint still bites.
Solar, storage and data centres wait years for a yes; reinforcement takes a decade. Every connection you can safely say yes to sooner is revenue for them, and headroom put to work for you.
Storms, faults, attacks. The bad days are decided in advance, by whether you know the capacity picture for the days ahead: which generation is available, and where the network is exposed.
Guiding the flexibility team's procurement across SSEN's constraint-managed zones in Scotland, refreshed hourly, up to 15 days ahead.
The same node-level foresight, run over scenarios instead of the live feed: what a new connection does to a feeder before anyone signs it.
Every zone forecast is graded daily against what actually happened. The operator sees our error before we do.
Engines forecast demand, generation and flexibility at every substation, feeder and constraint zone. A physics model of your real network turns those into flows, voltages, loading and margins ahead of time. The decision layer turns that into ranked actions and, where you want it, signals. It sits beside your ADMS, never inside it.
Of constraint foresight, per zone and per hour, with the responsible limit named: thermal, voltage or security.
Every forecast, refreshed every hour, at 15, 30 or 60-minute resolution, matched to your settlement.
Substations, feeders, grid supply points and constraint zones: one engine each.
Lead time. Days of warning, not minutes. Enough to buy the cheap fix instead of paying for the emergency one.
Named cause. Thermal, voltage or security, and which asset drives it. The conversation starts at the answer.
Confidence. Every event carries a probability. Your team triages honestly instead of treating every alarm the same.
And the window holds across runs: the same event, firming from forecast to forecast, is what earns the warning its trust. You watch it harden before you spend a cent on it.
Every forecast carries a calibrated confidence interval, and the physics carries it through: a constraint arrives as a probability, the recommended volume as a range. When confidence is high and policy allows, the decision automates; when it is not, a person decides. No team can hand-manage thousands of constraint calls a year; with confidence carried end to end, they do not have to. And the confidence is earned, not asserted: every interval is graded daily, so yesterday's 80% events should bind about eight times in ten, and you can check that they did.
Which actions clear the constraint, and with how many MW: ranked by engineering merit, no market data needed. Euros join wherever your market and tariff feeds connect: procurement versus curtailment, priced side by side.
Every forecast is a calibrated range, so procurement becomes a priced decision instead of a guess: cover the expected case cheaply, or pay a little more to be safe on the bad day.
One product, one contract, four levels, with the audit trail recording every step. Protection and sub-second functions never move.
Here is the constraint, and what it costs.
Curtail asset A by 3 MW from 14:00, or buy 5 MW in zone X.
Send the requirement to the flexibility market, or a dispatch instruction to a VPP.
Gridian Control: automatic curtailment and setpoints within playbooks you approved in advance, executed by the certified control layer.
Offer non-firm terms with a quantified profile ("about forty curtailed hours a year") instead of a decade-long wait in the queue. The connecting party sees exactly what they're signing; you see exactly what you're promising.
The same time-resolved headroom drives planning: it shows where reinforcement actually pays, and where flexibility covers the gap.
Thermal and voltage constraints flagged before they bind: limiting asset, timing, duration, severity.
DSOs buy local flexibility ahead of the constraint. TSOs size redispatch and balancing volumes from headroom.
Proposed connections evaluated against forecast conditions across the horizon, not one snapshot.
Where, when, why and how much: the causing constraint identified, lost energy quantified, the answer defensible.
Controllable DER volumes per location up to 10 days ahead, with a VPP interface to respond to dispatch.
Electrification and DER growth applied to your model over years: where reinforcement pays, where flexibility covers.
Gridian sits beside it, never inside it. Levels 1–3 need no OT installation at all; level 4 executes only through the certified control layer, within playbooks you approved.
Deployable in your tenant or fully on-premises; data can stay in the substation it came from. Engines run only inside signed Ogre AI runtimes: custody by design.
Gridian works with your existing ANM or DERMS: the predictive and pricing layer on top of it. No OT change, live in a pilot.
No ANM yet, or replacing one? We deliver it: forecasting, decisions and dispatch, with certified control partners wherever control is in scope.
Modular by API: forecasts, constraint events and headroom feed whichever ANM or ADMS you choose, today or later.
Hosted and operated by us: the fastest start, live in weeks. Nothing to install, nothing for your IT roadmap to absorb.
The same product inside your cloud tenant or on-premises, a live pattern with a 3.4-million-consumer utility. Engines never leave Ogre AI runtimes, wherever the runtime lives: custody by design.
Act days before the limit instead of paying compensation at it. And prove the counterfactual when generators ask.
Procure MW in the market ahead of time, at planned prices, not as an emergency at any price.
Non-firm offers with a quantified profile turn a decade of queue into revenue this year.
Time-resolved headroom shows where copper actually pays, and where flexibility covers the gap for less.
Your network and its history land on our side, and engines spin up for every node. Nothing gets installed on yours.
Constraint events for your zones, graded daily against what actually happens, visible to your whole team.
Hit rate, lead times and the value of every lever, quantified on your own history, reported honestly either way.
Pilots compare predicted against actual constraint events on your operator data over one to three months. Zones are added in weeks.
System imbalance every fifteen minutes, prosumer generation as a zone aggregate, and the fundamentals around them, from a single park to a whole country. For the desk online, or for the whole trading floor as a platform.
A system-imbalance and balancing-price engine per zone, refreshed every quarter hour, with the probability of the system being long or short.
Net feed-in is the curve that breaks zone forecasts. We forecast the prosumer aggregate per bidding zone as a product of its own: the number the settlement actually sees.
Watch the imbalance view firm up run by run inside the delivery hour. The drift is the signal, and it reaches you before the settlement does.
Alerts on surplus and deficit thresholds, run history, ⌘K everywhere, plus Excel, API and Python for everything else.
Sign and volume as they settle, the next hours as graded probabilities, the imbalance price on its own axis. Every probability is scored against what actually happened, so the desk knows exactly how much to trust the next one.
Sign and volume as they settle, every fifteen minutes. The next hours as P(surplus) per interval, with depth. The imbalance price in context, on its own axis. Alerts when the sign flips or the depth crosses your threshold, pushed to the desk before the settlement does it for you.
Net-metered feed-in, aggregated per bidding zone: the fastest-growing driver of imbalance, delivered as a forecast product like wind or solar, nowcast to ten days.
Zone-level load, wind, solar and net load as distributions, with flows, capacities and cleaned outages around them. Not the differentiators; the foundation under them, done properly.
System load, wind and solar by bidding zone, and the net (residual) load they imply: the number the price actually clears against.
Surplus flips, deficit depth and forecast revisions, pushed to the desk.
Cross-border flows and available capacity, per border and hour.
Unit and interconnector outages parsed, deduplicated and mapped to capacity, whatever the zone publishes them in.
Clean history to test any strategy: the same data our engines trained on.
Wherever your desk lives, including an app worth keeping open all day.
Fundamentals check: load, wind and solar for your zones. Where our view differs from the public one is where the opportunity is.
The distributions, side by side: how wrong could this be, and what does wrong cost at today's imbalance prices?
Alert: P(surplus) crossed your threshold for the evening block. Adjust intraday while the screen shows the system settling long, interval by interval.
Every probability we gave you, graded against what settled, so tomorrow's trust level is a number, not a feeling.
A single wrong-side imbalance interval avoided in a volatile week typically covers months of subscription. And volatile weeks are the new normal.
Trading on the whole distribution instead of the median compounds. The difference shows up in the month's P&L, not in any single trade.
Suppliers running per-meter engines, prosumers included, typically cut balancing costs by double digits across the book.
Pick your bidding zones and modules, pay online per seat, cancel any month. The app, Excel plugin, API and Python client ship with every seat; alerts and grading are the product, not add-ons.
The same engines as a platform: unlimited seats and API volume, custom zones and horizons, SSO and DPAs, data licensing for your own systems, and an agreement your procurement team can sign.
A zone goes live when its engines beat the local baselines on that zone's own history, and not a day before. Romania is first. Ask about your zone and we'll tell you exactly where it stands, either way.
Forecast the site, optimize battery, load and EV inside the connection limit: data centres, factories, industrial parks, campuses and prosumer communities. The grid connection is the new ceiling; foresight is how a site grows inside it. Advisory first; automated when trusted.
Data centres and factories are sized by their grid connection, and a bigger one takes years.
Prices swing hourly, imbalance every 15 minutes; flat consumption pays the worst of both.
Batteries, flexible processes, EV fleets and rooftop solar: installed, but not orchestrated.
Building systems execute rules; nobody on site forecasts tomorrow and plans against it.
Everything the site needs to know flows in; every asset behind the meter gets its schedule out. Continuously, locally, inside your limits.
Plan versus actual versus baseline, per asset, per day: exportable, and shown in the app where the savings counter earns its keep.
Where the network operator publishes signals (a constraint coming, headroom at your connection, a flexibility request), the site plans with them, whichever system they come from.
Deployable inside your own infrastructure when residency requires it: the same engines, running under your rules.
Energy Hub plans each site's day: what to run when, when the battery charges and discharges, how far load can flex. Always inside the connection limit, always priced against tariffs and markets.
The connection cap is never crossed: peaks shaved, penalties never paid.
Tariffs, day-ahead, imbalance: flexible load runs when prices are low.
Advisory schedules first; supervised through your management system; direct setpoints only when you switch it on.
The plan runs at whatever step your site does: 15-minute, 5-minute, down to 1-minute where control needs it, and it re-plans continuously as reality drifts from forecast.
Battery charged overnight on cheap hours; the day's schedule for load, storage and EVs is on the screen before the first coffee.
Solar under-delivers against plan. The schedule revises itself: flexible load slips an hour, the battery holds.
On-site generation peaks. Flexible processes run now, the battery tops up, and the site barely touches the grid.
Grid prices and site load climb together. The battery discharges on schedule; the connection limit is never even approached.
Plan versus actual versus what the old habit would have cost, on the screen every night, building the case for more autonomy.
The billed peak is a forecastable event. Shave it with the battery before it happens, not after it's on the invoice.
Arbitrage, peak shaving and backup readiness in one schedule. Most batteries do one job; a forecast-driven plan stacks all three without ever breaking the reserve.
A site that forecasts itself is a site its supplier prices better. Fewer surprises in the profile, fewer penalties passed through.
The cheapest megawatt is the one you don't have to apply for. Staying inside the limit defers the upgrade, and for a growing site, that's revenue.
Connections are the new constraint everywhere. A site that manages its energy with foresight gets what money can't buy quickly: headroom.
Compute growth capped by the connection: foresight converts directly into rack space, cooling into the cheap hours, and the UPS into an earning asset.
Process flexibility, on-site generation and storage, orchestrated against tariffs and imbalance instead of run on habit.
One connection, many tenants. The park plans as a whole: headroom allocated by forecast, the shared battery worked for everyone, and the growth question answered with data.
Hundreds of small, noisy prosumers, forecast individually and as a cluster, the way we already forecast millions of consumers for suppliers.
Tomorrow's schedule on your team's screen: when to charge, when to flex, what it saves. People execute; the system explains.
Schedules flow into your BMS or EMS; your operator confirms with one action. The audit trail records every step.
Setpoints to battery and flexible load within envelopes you approved in advance, switched on per asset, only when trusted.
Meters, tariffs and the connection limit. We see the site; we touch nothing.
Every day gets a plan and every plan gets graded: against what actually happened, and against what the site would have done anyway.
If the plan beats the habit, you keep it and start acting on it. If it doesn't, the numbers show you why.
The connection limit is not a preference in a dashboard; it is a hard constraint in the optimizer. Every plan is feasible by construction, every action is audited, and the day the utility asks, the record answers.
Bring a site with a battery, a connection limit and a bill worth shrinking, and help us build the product around it.
Twenty-five people, three offices, engines in production across seventeen countries. We hire people who want their work measured: accuracy published, shipped things live at real operators.
Everything else is packaging. We build one thing, five ways.
It is the only claim that matters, and it is graded every single day.
Our customers run critical infrastructure. We ship, we measure, and we stay accountable for what happens next.
Sovereign clouds, substations, grids under stress: the places that prove an infrastructure company.
Every application gets a reply. The process: a conversation, a practical exercise close to the real work, and a final round with the team you'd join. Typically two to three weeks end to end.
Tell us what's hardest to forecast. We'll usually come back with a proposal to measure ourselves against your current method, on your own history, with the result shown either way.
Forecasting is not a model problem. It's a scale problem. So we don't build models; we built an assembly line that builds, tests, stamps and ships them. One engine is born below every thirteen seconds, on repeat.
Register a solar plant, a substation, a portfolio of two million consumers or a country's power price, and a dedicated engine is built for it: fed by everything you have, from SCADA and meter telemetry to market data, tariffs, calendars and six calibrated weather sources, then measured, kept accurate and rebuilt when the world changes.
A plant, a substation, two million consumers, a market.
Data selected, weather calibrated, models chosen. Automatically.
Tested against history and stamped with its measured accuracy.
In our cloud, your tenant, or a computer in the substation.
Watched in service, rebuilt the moment accuracy slips.
Inside the model race, neural architectures compete with classical machine learning on every engine: LSTMs for the sequences, autoencoders hunting anomalies in the training data, quantile and generative ensembles turning point predictions into full distributions, all racing gradient-boosted trees that still win on many tabular problems and linear models that anchor stability. And the champion is rarely a single winner: engines ship as weighted ensembles of these (the Σw below), re-weighted as regimes shift.
The factory doesn't have a favourite. It has a scoreboard: architectures are chosen per engine, per horizon, by measured error on that asset's own reality, over ten years of weather archives and every production run ever scored. When fashion and measurement disagree, measurement ships.
Everything you have goes in (telemetry, market results, outage feeds, topology, tariffs, asset metadata), each cleaned, gap-filled, cross-checked and calibrated against outcomes. And above all: weather, the single biggest driver of accuracy.
No weather model wins everywhere, so we stopped asking one to. Our own weather-calibration AI (six providers in, one learned consensus out) learns which model to trust for which variable, in which range, at which horizon, per site, and re-learns it on every provider update. Trained on years of our own forecast-versus-actual history: not licensed, not replicable from a datasheet.
An Ogre AI engine is a computational tree: hundreds of sub-engines aggregated into one number with a band around it. Each part is chosen by the factory, per engine, never hand-picked.
Meters, SCADA, market feeds, and weather from six sources, calibrated to the asset's own geography and equipment.
Outliers removed, sub-optimal operating periods detected and reported, so engines never learn from broken behavior.
From base signals to selected, transformed features, with correlation, importance scoring and domain knowledge deciding what stays.
From linear models and gradient-boosted trees to LSTMs and autoencoders, raced against each other, per engine, per horizon.
Sub-engines ensembled into one forecast, with distributions, not just points, so every number carries its uncertainty.
Smoothing and constraints from the asset's technical reality, applied last, so forecasts respect the physics they describe.
Forecasting is half the machine. The other half turns calibrated ranges into actions. It is what makes Gridian a grid product and Energy Hub a site product, instead of two more dashboards.
Load-flow on your CIM/GIS network model, driven by every node's forecast. Constraints found, headroom computed, the limiting asset named.
Site days planned under tariffs, connection limits and battery physics. Procurement sized from calibrated ranges, with cost functions a CFO recognizes.
Advise, recommend, signal or close the loop. Policies approved in advance, per zone and per asset, with certified partners wherever control is in scope.
Not just forecasts: plan versus actual versus baseline, per action. The counterfactual is kept, so the value claim survives an audit.
A point forecast can only be right or wrong. A calibrated band lets you price the trade-off: cover the expected case cheaply, or pay a little more to be safe on the bad day, knowing exactly what that insurance costs.
The same mathematics sizes flexibility for Gridian, plans batteries for Energy Hub, and prices risk for the Markets desk. One engine room, three products.
It also keeps everyone honest. A safety margin hides in a spreadsheet; a coverage level is a number someone chose, wrote down, and can defend a year later.
Every action carries its whole ancestry: the forecast run, the engine version, the physics run, the option chosen, the policy that allowed it, and the graded outcome. Exportable for a regulator in one click.
The first question every infrastructure buyer asks, answered the same way everywhere in the stack.
Researchers race variants of any station (features, models, aggregation) across many engines in parallel, and the winners are promoted.
Intermediate results are cached, so changing step seven never recomputes steps one to six. Research moves at the speed of the idea.
Any engine can be replayed exactly as it ran in production, with what it knew and when it knew it, so unexpected behavior becomes a report, not a mystery.
And the bets we are placing now, so that when today's edge becomes table stakes, the line has already moved:
One model pretrained across millions of curves. Every new asset starts smart and fine-tunes from there, instead of learning from zero.
Decisions that learn from their own consequences: reinforcement learning proven in simulation before it ever touches a switch.
Engines that read the grid diagram, not just the timeseries: graph models that know which feeder feeds whom, and what fails together.
AI weather models are starting to beat physics on some horizons. Our calibration AI folds each one in the day it wins, per region, per regime.
A pilot registers your hardest curve, builds its engine, and grades it against your current method, on your own history.
The engines behind Forecast and Gridian, entering two new rooms: the trading desk and the site behind the meter. In early access now and available from the end of 2026, priced for the people who bet with us. The deep dive happens in a conversation, not on a page.
System imbalance every fifteen minutes, prosumer generation as a zone aggregate the settlement actually sees, and the fundamentals around them, from a single park to a whole country. All of it as distributions, with the cost of being wrong made visible. Bought online, at a published price, from the end of 2026.
Forecast-native optimization inside the connection limit: the battery's day, the flexible load, the on-site generation, planned together for data centres, industry and energy communities. Advisory first. Autonomy is earned, action by audited action.
Four engagements across the value chain (a DSO group, a retailer and two national TSOs), each still running. Figures and quotes as shared in our client materials.
Three local DSOs needed losses down urgently, with price volatility and balancing costs hitting the business and pandemic-era history making clean training data scarce.
Demand and technical-losses forecasting across all three DSOs with a customized reporting platform, with engines retrained continuously as conditions normalized, graded monthly against settlement.
“The AI application we implemented for reducing technical losses has revolutionized our operations. Our efficiency has skyrocketed.”
A supplier with 3.4 million consumers needed accurate consumption forecasts to reduce balancing costs and boost profitability, on a tight timeline, with the war in Ukraine next door and unfavourable regulations to work around.
The full stack: data integrations and data management, the AI forecast engine and the Ogre reporting tool, implemented for all 3.4 million consumers in just a few months, inside the client’s own Azure cloud.
“The AI-driven solution we adopted to address profit margin problems has been transformative for our energy company. The impressive efficiency gains and heightened sustainability practices have placed us at the forefront of innovation.”
Elia set more than twenty of Europe’s leading forecasters against each other on Belgian offshore wind and solar: blind, no access to peer results, judged purely on accuracy.
Ogre AI forecasts delivered into Elia’s multi-provider setup, graded continuously against every competitor. The benchmark still runs, and so do we.
“The setup is blind and performance-based — making accuracy the sole differentiator.”
More than 10 GW of scattered wind capacity, much of it unmetered, plus local grid limitations, and a national control room that needs one dependable picture of it all.
Real-time generation forecasts for every wind farm in the country and its aggregation areas, feeding TSO operations around the clock: responsibility carried daily, at national scale.
“In order to manage the backbone transmission system of a country, you need reliable and accurate systems to support grid operations. We have implemented the Ogre AI forecasting system for all the wind farms in the country.”
Every engagement starts the same way ours end: measured.
Last updated: 30 September 2026
Who we are. This website is operated by Ogre AI ("Ogre", "we"), 6 Gara Herastrau, Globalworth Square, 3rd floor, 020334 Bucharest, Romania, with offices at 21 Ellis Street, Knightsbridge, London SW1X 9AL, United Kingdom, and Toho Hibiya Promenade Building, Yurakucho, Tokyo, Japan. Ogre AI is the controller of the personal data collected through this website. For anything related to your data, write to office@ogre.ai.
Enquiries. When you use the contact form or write to us, we process the details you give us (your name, work email, company, the product you are interested in and your message) to answer you and, where relevant, to discuss a potential engagement. The legal basis is our legitimate interest in responding to business enquiries (Art. 6(1)(f) GDPR) and, where discussions progress, steps taken at your request before entering into a contract (Art. 6(1)(b) GDPR).
Job applications. When you apply for a role, we process what you send us (your name, email, phone number, location, links such as LinkedIn or a portfolio, your earliest start date, your note and your CV) to assess your application and run the recruitment process, which is a step taken at your request before a possible employment contract (Art. 6(1)(b) GDPR). If you tick "Keep my application on file for future roles", we keep it on the basis of your consent (Art. 6(1)(a) GDPR), which you can withdraw at any time. We do not make hiring decisions by automated means.
Technical data. When you visit the site or submit a form, our servers process your IP address and basic request information (such as browser type and time of access). We use it to deliver the site, keep it secure and prevent spam, for example by limiting how many forms can be sent from one address in a short time. The IP address is also included in the notification we receive for each submission, so we can recognise abuse. The legal basis is our legitimate interest in running a secure website (Art. 6(1)(f) GDPR).
Cookies. This website does not set cookies and does not use analytics or advertising trackers. If we introduce analytics, it will be privacy-respecting and this policy will say so before it happens.
Third-party services on this site. The site loads its typeface from Google Fonts, which means your browser connects to Google's servers and shares your IP address with Google when a page loads. Forms are protected by Cloudflare Turnstile, which processes information about your browser and device to tell people apart from automated bots, without setting cookies. Both providers act under their own privacy terms.
Who receives your data. Your data is accessible only to the Ogre AI team members who need it: the commercial team for enquiries and the hiring team for applications. It is processed on our behalf by the providers that run our infrastructure: Google (Google Workspace, which delivers form submissions to our inboxes), our website hosting provider (GoDaddy) and Cloudflare (form protection). We do not sell personal data or share it for advertising. Where a provider processes data outside the European Economic Area, we rely on an adequacy decision, such as the EU-US Data Privacy Framework, or on standard contractual clauses.
How long we keep it. Enquiry correspondence is kept for as long as needed to handle the enquiry and any resulting relationship, and no longer than three years after the last contact unless a contract or a legal obligation requires otherwise. Applications are deleted within six months of the recruitment process ending. If you asked us to keep your application on file, we keep it for up to two years, or until you withdraw your consent if sooner. Server and security logs are kept on a rolling basis for as long as needed to operate and protect the site, and are then overwritten.
Security. The site is served over an encrypted connection, form submissions are delivered only to our own mailboxes, and access to personal data is limited to the people who need it.
Your rights. Under the GDPR (and, for the United Kingdom, the UK GDPR) you may request access to your personal data, its rectification or erasure, restriction of or objection to its processing, and its portability, and you may withdraw consent at any time without affecting processing that happened before. To exercise any of these rights, write to office@ogre.ai; we respond within one month. You can also lodge a complaint with a supervisory authority: in Romania, the National Supervisory Authority for Personal Data Processing (ANSPDCP, dataprotection.ro); in the United Kingdom, the Information Commissioner's Office (ico.org.uk).
Changes. We will update this page when our practices change. The date at the top shows the latest revision.
Last updated: 30 September 2026
Scope. These terms govern your use of the ogre.ai website. They do not govern the Ogre AI products and services themselves. Ogre AI Forecast, Gridian, Ogre Markets, Energy Hub and related offerings are provided under separate written agreements with their respective customers, and nothing on this website forms part of those agreements or constitutes an offer.
Content. The information on this website is provided for general information. Figures, availability, product descriptions and performance references may change, and we may update or remove content at any time without notice. Nothing on this site constitutes trading, investment, engineering or legal advice, and forecasts or illustrative visualisations shown here are demonstrations, not data to act on.
Intellectual property. The Ogre AI name, the Ogre AI mark, the product names Ogre AI Forecast, Gridian, Ogre Markets and Energy Hub, and the content, design and illustrations of this website are the property of Ogre AI or its licensors. Third-party names and marks, including those of customers and partners referenced on this site, belong to their respective owners and are used to describe real relationships, not to imply endorsement beyond them. You may not reproduce or reuse site content commercially without our written permission.
Acceptable use. You agree not to misuse this website: no attempts to breach its security, scrape it at scale, misrepresent your identity in forms, or use it to send unlawful or abusive content.
Liability. The website is provided "as is". To the maximum extent permitted by law, Ogre AI is not liable for damages arising from use of, or inability to use, this website or reliance on its content. Nothing in these terms excludes liability that cannot be excluded under applicable law.
Links. Links to third-party sites are provided for convenience; we are not responsible for their content or practices.
Forms and applications. When you send an enquiry or apply for a role, you confirm that the information you provide is accurate and that you are entitled to share it. How we handle that information is described in our privacy policy.
Governing law. These terms are governed by the laws of Romania, and disputes relating to this website are subject to the jurisdiction of the courts of Bucharest, without prejudice to mandatory consumer protections that may apply to you.
Changes. We may update these terms from time to time. The date at the top shows the latest revision, and continued use of the website means you accept the current version.
Contact. Ogre AI, 6 Gara Herastrau, Globalworth Square, 3rd floor, 020334 Bucharest, Romania · office@ogre.ai
An agent can get a forecast for anything Ogre AI already forecasts, or register something new with its own data and have a dedicated, retraining engine built for it. Every answer carries the engine's live accuracy, so the agent knows how much to trust it.
Get the numbers for an entity and horizon.
Register an entity with data and get an engine.
Accuracy, backtests and drift for any engine.
What if the weather run, capacity or horizon changed.