The operational data platform for AI agents and analytics
Your operation already produces the data. What is missing is the connection: readings that know which pump they came from, events that know which process they interrupted, numbers you can trace back to a source. That connected model is what your people and your AI agents both work from, and building it is what this platform is for.
The product
Real screens, not mock-ups
The console your team would open every morning, exactly as it ships.
-
Your operation as one connected map Assets, the work they do, and the data they produce — laid out as one picture you can follow, instead of five systems you have to reconcile by hand. Every derived value keeps its lineage, so any number traces back to the signals behind it. -
The plant in its own language Wells, manifolds, separators, pumps and the pipes between them, modelled with the relationships your engineers already use. Pick any node and follow what it feeds, what monitors it, and what stops if it stops. -
Measurements as they arrive Follow a signal as it comes in, compare it against others, and move between the last few minutes and the last year without waiting for a report to build. Each one stays attached to the asset it came from, so the context travels with it. -
Machine learning that keeps watch Anomaly detection learns what normal looks like for each signal, so you are not maintaining long lists of fixed thresholds. Build AI agents that monitor your measurements around the clock and raise an event the moment something drifts out of pattern. -
Every alarm points back at an asset An anomaly arrives with the model that raised it, the threshold it crossed and the exact sensors it concerns. The question after "something happened" is already answered.
The feature set
What you get, feature by feature
The rest of this page tells the story. This is the list, for the reader comparing platforms with a spreadsheet open. Two entries are not finished yet, and say so.
One model of the whole operation
An ontology mapping framework connects your assets, the functions they perform and the business knowledge around them into one model. Every reading arrives with its meaning attached, so nobody has to ask what a number refers to.
Explore the knowledge graphLive data for everyone
Anyone you give access opens the platform and reads the operation as it is right now. Nobody files a ticket with IT to get at a number, and nobody waits for an export.
Governance for people and agents alike
Access is decided by policy, dataset by dataset, and the same policy governs a person and an AI agent. Nobody grants an agent blanket access just to get it working, and when someone asks who could see what, the answer is already on record.
Zoom from a year to a millisecond
Drag across a chart and move from twelve months of history down to the individual raw datapoints, live, with aggregation running over billions of points as you go.
History that stays affordable
Long history is what makes an AI model worth training, and it is the first thing most operations throw away, because storage is expensive. Time-series data compresses around a hundred times on the way in, using the best compression the field has produced, so you can afford to keep the years your models will need.
Connectors that speak to the plant floor
Modbus and OPC UA connectors bring readings straight from your equipment into the model. Getting a new signal in is configuration work, not an integration project.
See how integrations workAn SDK for the team that builds
Everything the console does, code can do. The SDK is for your developers, whether they write the code themselves or put AI agents to work on it, and it is public along with its documentation, so they can judge it before anyone signs anything.
Send your developers to the SDK docsAgents connect the way people do
AI agents plug in over MCP, the open standard assistants use to reach tools, behind the same permissions as everyone else.
See what that looks like on a real pumpYour cloud, your datacenter, or ours
Run it where your rules say it must run. Sovereignty is a feature here, not an afterthought: your data can live on hardware we own in Norway, or never leave your building at all.
Read about the sovereign cloudA map of your data flows Coming soon
A data-flow framework with a visual map, so you can see where every stream starts, what transforms it and where it lands. It is being built now, and it ships when it is good.
Follow it on the roadmapAgent orchestration Coming soon
Autonomous workflows carried out by agents that watch, decide and act on the events you define, with a person approving exactly where you decide one should.
What it leads to
Six things change when the data lives in one place
No feature on that list matters on its own. What matters is what they add up to once the whole organisation reads the operation from the same place.
People stop hunting for information
The answer lives in the platform, not with whoever answered it last time.
Less unplanned downtime
Drift gets spotted while it is still a maintenance task, not yet a shutdown.
Maintenance money goes further
Spending follows the measured condition of the equipment instead of the age of the schedule.
Turnarounds get faster and cheaper
Planning starts from the actual state of every asset, so the scope holds when the work begins.
Decisions rest on the same numbers
When two reports disagree, that is a data-quality finding to fix, not a meeting to hold.
AI becomes usable
An agent reading a governed model gives answers you can trace, and that is the difference between a demo and a tool.
Return on investment reported in Forrester Total Economic Impact studies of two operational data platforms in this category: 400% with 21.6 million dollars in added net present value in one study, 315% with 262 million dollars in the other, each for a composite asset-heavy organisation.
Those numbers are not ours, and that is the point. The studies behind them were commissioned by the incumbents that defined this category, and they measured platforms that cost many times what ours does. We read them the way you should: as independent evidence that an operational data platform pays for itself in asset-heavy operations.
Our own promise is plainer. Put the operation in one live model and the pace of everything built on top of it picks up: new questions get answered the week they are asked, and trying an idea stops being a project. That is what we mean when we say this platform will raise your organisation's pace of innovation.
What it does
Data you can trust, working on its own
Three things decide whether a data platform earns its place. Whether whoever is reading it, person or agent, can trust what they are looking at. Whether it handles the routine work without being asked. And whether the business hears about a problem while there is still time to act.
Governance built in
Your data is organised into datasets — the parts of the operation that belong together. Access is granted per dataset and by role, so every person and every system sees exactly what they should and nothing more. How a dataset is meant to be handled is recorded against the dataset itself, next to the data it describes, rather than in a document nobody opens.
- Access decided per dataset, not per spreadsheet
- Handling rules kept where the data lives
- A straight answer when someone asks who could see what
Intelligence that keeps working
Put a function on a measurement and it goes to work on every new reading — watching for values drifting outside their normal range, projecting where a trend is heading, turning a pattern into something the business can act on. You are told when something looks wrong instead of waiting for someone to notice it on a dashboard.
- The routine watching runs by itself
- Findings arrive as events the business can act on
- Your people spend their time on the exceptions
An operation that reacts
A limit crossed, a state change, a prediction worth knowing about — each becomes an event the moment it happens, and the systems that care about it are told straight away. Decisions that used to wait for the morning report get made while they can still change the outcome.
- Things are known when they happen, not the next day
- Systems can act without a person in the middle
- Less time gathering, more time deciding
Who reads it
Your people and your agents read the same model.
An agent that has to be told what a number means will get it wrong eventually. DataHub gives agents the same connected picture your engineers work from, through the same door, with the same permissions. Adopting agents does not open a second way into your data. Once building is cheap, the cost of integrations is the constraint that is left.
One door, one permission model
Agents connect over a standard interface, behind the same security checks as everyone else. No service account with blanket access, and no separate login for agents.
- An agent inherits the permissions of whoever it acts for
- Read-only stays read-only, because the token says so
- Customer separation runs through the same code as the console
Tools that work on your own model
An agent can search your equipment, walk from a reading to the process it serves, pull history between two dates, and write back what it finds.
- Equipment, relationships, time series and events
- Record an event or append a reading, not only read
- The same model your reports are built on
Answers sized for a machine
Search and browse results are trimmed to what matters, so an agent spends its attention on the few records that decide the answer rather than on everything it happened to look at.
- Empty fields are never sent
- Lists return summaries, full detail on request
- Timestamps read the same as over the API
What that looks like on one pump
None of the five steps below live in the same system. That is the whole difficulty, and it is why an agent without a connected model gets stuck at step two.
The spare is linked to the pump, the technician's qualification to the equipment, the compliance obligation to that same asset. That is what lets an agent get all five right. DataHub is not your purchasing system or your permit system, it is the model that tells the agent which pump, which part, which technician and which obligation belong together, so the calls it makes into those systems are the right ones. Where a person approves is yours to decide when you build the agent, not something the platform decides for you.
Data quality
Six dimensions, always in check
Every number is checked against six dimensions as it arrives, so a figure that reaches a report, a dashboard or an agent has already earned its place.
At the heart of our data-platform lies a revolutionary approach to data organization. Knowledge Graphs
Getting started
How this starts, and how it keeps going
There is no eighteen-month rollout hiding at the bottom of this page. We meet and learn about each other.
You call, we listen
We begin by learning your business: what you run, the problems that cost you money today, and where you want the operation to be in five years.
Your engineers, our computer scientists
People who know data platforms working next to people who know the plant is a powerful combination for solving problems and moving forward. Together they find the low-hanging fruit, the places where the platform pays for itself first, and start there. We call that work product discovery, and it has a page of its own.
How product discovery worksYou keep building
This is not a one-shot solution, and we will not pretend otherwise. The model grows one process at a time, and every process you add makes the next one cheaper and the answers richer.
None of it has to happen at once. But compounding is the point: the operation that starts this year is answering questions next year that its competitors have not yet learned to ask. That edge cannot be bought later, it has to be built up.
Frequently asked
The questions that decide whether this is worth it
What does this give us that our reporting does not?
Reports answer the questions someone already thought to ask. An operational data platform holds the model of the operation itself: the assets, the processes and how they relate, with measurements and events attached to it. A new question is then answered from what is already there, instead of becoming another data project.
Can we let AI agents work against our operational data?
That is what the model is for. An agent reading a raw export has to guess what a column means, and it will guess confidently when it is wrong. An agent reading the model is told what each asset is, what it measures and what it connects to, so its answer can be traced back to the same source a person would check. Your people and your agents work from one description of the operation rather than two.
How do we know the numbers are right?
Six dimensions are checked continuously: accuracy, completeness, consistency, timeliness, uniqueness and validity. A missing measurement shows up as missing rather than quietly lowering an average, an asset recorded twice is caught before two teams quote different figures for it, and a value that cannot be true is flagged where it enters rather than in a report months later.
Will it keep up with our data volumes?
The engine underneath reads over 10 billion rows per second in our setup, and time-series data compresses around 100 times, so ten years of second-level history is a storage decision rather than a budget argument. For most operations throughput is not the real constraint; how long it takes to get the operation modeled in the first place is.
What does it take to run?
Less than most teams expect. Run it yourself in your own datacenter or cloud, or have us host it on hardware we own and operate. Either way the code is open under AGPL-3.0, so what you come to depend on can be read, scrutinized and kept running by someone other than us.
What does it cost?
The software is open source under AGPL-3.0, so running it yourself costs hardware and your team's time, not licences. What we charge for is the work around it: managed hosting on infrastructure we own, support, and fixed-price pilots scoped against a KPI you name. Because our cost structure is small and the hardware is our own, the total routinely lands around a tenth of what the category's enterprise platforms quote.
Do we have to replace the systems we already run?
No. Control systems are not touched, and a process historian that works stays where it is. The platform is the layer above them: it reads from what you already run, connects those readings into one model, and gives people and agents one place to ask. Adding a layer is a smaller job than a migration, and unlike a migration it can be undone.
Where does our data actually live?
Wherever you decide. Run everything in your own datacenter or your own cloud account, air-gapped if your rules require it, or have us host it on hardware we own and operate in Norway, under Norwegian jurisdiction. And because the code is open under AGPL-3.0, where your data flows is something your engineers can verify themselves instead of taking on faith.
How do we start without committing everything to it?
Take a single process that already causes arguments about the numbers, and model that one. The platform is open source, so you can load real data and watch it work before there is a contract to sign. Company-wide rollouts are where projects like this usually stall, which is why we advise against starting with one.