Context graph vs knowledge graph vs ontology
An ontology defines the kinds of thing, a knowledge graph fills them in, a context graph is the slice one task needs. Side by side, with a manufacturing example.
An ontology defines the kinds of thing in a business and how they relate. A knowledge graph is that ontology filled in with the real customers, orders and products. A context graph is the slice of that graph one task needs, plus current state and the decisions already made. They are three layers of one stack, not three rival tools.
The terms get used loosely, so here they are side by side, with one example from a manufacturer that quotes and makes parts to order. The example is an illustration, not a client case.
The three side by side
| Ontology | Knowledge graph | Context graph | |
|---|---|---|---|
| What it is | The definitions | The populated map | The working slice for one task |
| What it holds | Kinds of thing, how they relate and what may be done with each | The particular customers, quotes, orders, products and machines, and how they connect | The things one task touches, their current state and past decisions about them |
| Scope | The domain you model | Everything the ontology covers | One piece of work |
| Changes when | The business changes what it does | A fact changes, such as a new order or a new contact | Every time the task runs |
| In the manufacturer example | An Order refers to a Quote and needs Production steps | This customer's order refers to last month's quote and needs cutting and assembly | For confirming that order: the customer's terms, the quote, the routing, the machine queue and the last exception approved for this customer |
| Think of it as | The legend | The map drawn with it | The route for one trip |
A worked example: one order through all three
A customer emails an order. The ontology says what an Order is, that it refers to a Quote and needs Production steps, and that an agent may check it and draft the confirmation but may not send it. The knowledge graph resolves the sender to one customer record, whether that customer appears in the ERP, the CRM or an old spreadsheet, and finds the quote the order refers to.
The context graph is what the agent is handed for this one task: the customer's terms, the quote, the routing for each part, the current machine queue and the last exception approved for this customer, with a source on each. If everything matches, the agent drafts the confirmation. If a price is below the quote or a date clashes with the queue, it flags the line and a person decides. That decision is saved as a trace, so the next exception has a precedent.
Where a world model fits
A world model for knowledge work is the maintained version of the middle layer: the knowledge graph kept current, with time and a source on every fact and what the company knows attached to each thing. Context graphs are drawn from it. The difference is covered in context graph vs world model.
Which one do you need first?
All three, in that order, but for one piece of work at a time. Write the ontology for that work, fill the graph from the sources it depends on, then let each task draw its context graph. Trying to model the whole company first slows the first result down. The steps for the first layer are in how to build an ontology for AI agents.
Questions people ask
Is a knowledge graph the same as an ontology?
No. The ontology says a Customer can place an Order. The knowledge graph holds the particular customers and their orders. One is the legend, the other is the map drawn with it. More in what is an ontology for AI agents.
Is a context graph just a smaller knowledge graph?
Partly. It is cut to one task, and it adds current state, history and the decisions made about those things, with their reasoning. More in what is a context graph for AI agents.
Do AI agents need all three?
An agent that acts in a business needs to know what things are, which particular things exist and what matters for the task in front of it. Without that, it rebuilds the picture from tools on every run and fills the gaps with what is typical.
Where does Ortelian fit?
Ortelian is a platform plus a forward deployed team. We build the ontology, the world model and the context agents work from with your team during a deployment, on a platform hosted in the EU. The world model, the IP built on it and the data stay with the client.
