A knowledge graph connects what your systems keep apart
Your systems are full of nouns. Pumps, sensors, work orders, invoices, reported figures, each stored in its own table, in its own system.
The verbs are missing: which sensor monitors which pump, which pump serves which process, which process feeds the figure someone reports. A knowledge graph stores the verbs as data, and stored verbs are what make data intelligent.
What it is
One connected map, not another silo
IntelliStream connects your structured and unstructured sources into one shared model, carries context across every department, and makes the relationships visible, so bottlenecks, patterns, and impact across systems are traceable instead of buried.
In industry this map has a name, the industrial knowledge graph. It is the part of an operational data platform that knows how your operation fits together: which pump feeds which process, which process feeds which reported figure, and who answers for it.
A worked example
A thing, a verb, another thing
Pump P-101 serves separator train A. That is a sentence, and it is also a knowledge graph, the smallest one there is. The nouns are equipment. The verb is the relationship between them, stored as data you can follow.
Sentences chain. The sensor monitors the pump, the pump serves the train, the train feeds the export figure. With enough of them, questions that used to take a meeting become queries:
How it works
Two building blocks. That's the whole model.
Nodes are the nouns. Relationships are the verbs. Everything else is a property hung on one or the other, and that small vocabulary is enough to describe an entire facility.
Nodes are the things your operation is made of:
- Assets: pumps, compressors, sensors, facilities.
- Processes: separation, metering, maintenance, reporting.
- People: operators, engineers, vendors, whoever answers for a figure.
Each node carries properties, the details that describe it: a tag number, a commissioning date, a status.
Relationships are the verbs that connect them, each one stored as data:
- A sensor monitors a pump.
- A pump serves a separator train.
- A train feeds a reported figure.
Relationships carry properties too, a start date, a flow direction, so the context sits on the connection itself.
Put to work
Your operation is already a graph
A pressure reading belongs to a pump. The pump serves a process. The process feeds a figure someone reports to a regulator. Your systems store all of this in separate tables. Asset dependency mapping puts the connections back, so the answers are there before the questions get urgent.
Trace every reported figure
When an auditor asks where a number came from, follow it back through every calculation to the raw signals it was built from, with data-quality flags intact. Four hours of manual cross-checking becomes minutes.
Investigate failures in context
When a component fails, the graph shows what sits upstream and downstream: the equipment it serves, the processes at risk, and the people who need to know. Not five browser tabs and a senior engineer's memory.
Onboard new sources once
Connect a new data source to the model once, and every report and dashboard downstream inherits its context. Adding a system stops meaning rewriting everything that sits on top of it.
Where the graph comes from
Start from the drawings you already have
Every facility already has its knowledge graph on paper: the P&ID, the piping and instrumentation diagram that shows every pump, valve, and sensor, and how they connect. IntelliStream builds the graph from that sheet together with you: the platform reads, you confirm.
The platform reads the sheet
Upload a P&ID and the platform finds the tags on it, each with its exact place on the sheet. Drawings produced by CAD read directly; scanned archives can go through OCR.
You confirm the matches
The platform checks every tag against the graph, and you decide the rest: confirm a match, add what is missing, and see both gaps, tags on the sheet your systems do not know, and equipment in the graph the sheet never shows. Nothing enters the graph as a guess.
The drawing comes alive
A confirmed tag is a node in the graph. The sheet stops being a static PDF: click a tag and the live reading stands right on the drawing, in the place your engineers already look.
Why now
Your people can read the graph. So can your AI agents.
Readable across the operation
A graph reads like the diagrams your engineers already draw on whiteboards: things, with labelled lines between them. Operators, accountants, and auditors can follow the same picture, no query language required.
The context agents are missing
In a demo, an AI agent can answer from whatever it happens to reach. In production it has to know which asset list is current, what unit a reading is in, and what depends on the pump it is reasoning about. Today that knowledge lives in people's heads. The graph is where it gets written down, in a form an agent can query and cite.
Where it fits
Built into the operational data platform
A knowledge graph on its own is a map with nothing on it. In IntelliStream the graph is built in and already connected to everything else the platform holds: time-series storage that compresses industrial signals around 100x, live event streams, and lineage on every derived value.
Common questions
Straight answers to common questions
What is an industrial knowledge graph?
A connected model of your physical assets, the processes they drive, and the business context that gives their data meaning. Sensor readings, events, and documents attach to the equipment they describe, so a question about one part of your operation can follow real relationships instead of joins across five systems.
How is a knowledge graph different from a data warehouse?
A warehouse stores tables and answers the questions it was designed for. A knowledge graph stores relationships, so it also answers the questions nobody predicted, like which reported figures depend on one failing pump. Most operations need both. The graph carries the context; the warehouse carries the history.
Do we have to model everything up front?
No. Start with one plant area or one nagging question, and grow the model as new questions arrive. New assets and relationships integrate into the existing structure without a rebuild, which is exactly what a graph is for.
Is the knowledge graph a separate product?
No. In IntelliStream it is a built-in layer of the operational data platform, connected to time-series storage, events, and lineage from day one. The whole platform is open source under AGPL-3.0, so your team can read the code before you commit to anything.
What is asset dependency mapping?
The part of the model that records what depends on what: which equipment serves which process, which processes feed which reported figures, and which teams own them. It is how the graph turns a component failure into a list of consequences instead of a guessing game.
Can we run it on our own infrastructure?
Yes. Self-host it in your cloud, on-prem, or air-gapped. Or let us run it on our own hardware in Stavanger, Norway, more than 50 servers with 2 PB of disk, operated by the same Norwegian engineering team that builds the platform.
Get in touch
Curious what your data looks like as a graph?
Create a playground tenant, load your own data, and see the graph for yourself. A working environment in minutes, and no sales call to get there.