# Ontologies

> Ontology design from Boltcliq: objects, links and governed actions that give people and AI agents one shared model of the business, with worked examples for banks and farm cooperatives.

Canonical URL: https://www.boltcliq.com/solutions/ontologies
Part of: https://www.boltcliq.com/llms.txt

The shared model of your business that lets humans and agents reason about it and act on it safely.
- An ontology says what exists in your organization, how it connects and what can be done with it. It turns raw data into something an agent can understand and act on without guessing.
- Objects: Customers, assets, shipments, accounts, patients
- Links: How the objects relate in the real business
- Actions: Governed operations people and agents may perform
- Defined once: Used by dashboards, applications and agents alike
### What we do
- Every organization already has an ontology. It lives in the heads of experienced staff, in the column names of old databases and in the way people argue about what counts as an active customer. Writing it down as a model changes what you can do. Reports agree, applications share definitions and agents have a map instead of a pile of tables.
- An ontology has three parts. Objects are the things the business cares about, each with properties. Links are the relationships between objects, such as a borrower holds an account or a farmer delivers a lot. Actions are the operations that can change the state of the world, such as approve a loan or reject a delivery, each with rules about who may do it and when. Together they describe how the business really works, not how a schema happened to be built.
- We design ontologies with the people who run the process. Modelling is workshop and review work as much as technical work. Our engineers draw the model with your domain experts, test it against real questions, map it to the source systems and put it under version control. The result is the centre of everything else we build, from analytics to agents. Read our point of view on the ontology layer for the longer argument.
### An example: a farm cooperative
- To make this concrete, here is a small ontology for a farm cooperative. It is an illustration of the method, not a customer deployment.
- Objects: Farmer, Farm Plot, Delivery, Lot, Grade, Warehouse, Buyer, Payment and Input Loan. Each has properties such as location, crop, weight, moisture and date.
- Links: a Farmer owns Farm Plots, makes Deliveries and holds Input Loans. A Delivery becomes part of a Lot. A Lot is stored in a Warehouse, sold to a Buyer and settled by Payments. Actions: Grade Delivery, Merge into Lot, Release Payment and Flag Quality Dispute, each with an approver and a record.
- Objects with clear properties
- Links that match how the cooperative works
- Actions with owners, rules and an audit record
### Capabilities
**Object modelling**
- We identify the nouns of your business and define each as an object with properties, identifiers and an owner. We resolve duplicates, such as one customer with three IDs, by matching and by agreeing a golden source. Objects are modelled once and reused everywhere.
- Workshops with domain experts
- Identity resolution across systems
- Property definitions and units
- Customer means one thing, everywhere.
**Links and business logic**
- We capture how objects relate and the rules that govern them, such as a loan cannot exceed an approved limit or a lot must be graded before sale. Links are typed and directional. Rules are written in a form that people can read and systems can enforce.
- Typed, directional relationships
- Constraints and derived properties
- Rules readable by non-engineers
- The value is in the connections, not the tables.
**Governed actions**
- We define the actions that can change the world and attach permissions, validation and approval to each. An agent that wants to approve a payment calls the action. The action checks the rules and the actor, and writes a record. Nothing happens by writing directly to a table.
- Actions with validation and approval
- Permission checks per actor
- A record of every execution
- Safe automation begins with safe verbs.
**Mapping to source systems**
- We map each object and property to the systems that hold it, including core banking platforms, ERPs, spreadsheets, mobile money and scanned documents. Lineage records where each value came from. When a source changes, the mapping is updated and the model stays stable.
- Source mapping for databases, APIs and files
- Lineage for every property
- Stable model over changing systems
- The model should outlive any one system.
**Ontology governance**
- We set up how the model changes: proposals, review, testing, promotion across environments and deprecation. Owners are named for each object. Changes are versioned so applications and agents are not surprised.
- Proposal, review and promotion flow
- Named owners per object
- Versioning and deprecation
- A model nobody can change safely will be bypassed.
**Ontology for agents**
- We expose the ontology to agents through APIs and tool interfaces, so an agent can discover objects, query them and call actions within its permissions. Prompts and tools refer to the same definitions that dashboards use. Answers cite the objects and sources behind them.
- Query and action interfaces for agents
- Permissions per agent identity
- Answers that cite objects and sources
- Agents are more reliable when they share your vocabulary.
### Where it applies
- Banks
- A bank has separate views of the same customer in core banking, cards, mobile money and collections, so risk and service teams disagree.
- The ontology defines Customer, Account, Facility, Transaction and Case, with links and rules for limits, exposure and review. Reports and agents use it.
- The head of risk owns definitions of exposure and limit. Credit officers decide on exceptions.
- Farm cooperatives
- A cooperative tracks deliveries on paper and in spreadsheets, and disputes over weight, grade and payment are common.
- The ontology links Farmer, Plot, Delivery, Lot and Payment. Actions such as Grade Delivery and Release Payment are recorded and traceable.
- The cooperative manager and the farmer committee decide on disputes and payment rules.
- Healthcare
- Patients move between clinics and labs, and each holds a partial record under a different identifier.
- The ontology models Patient, Encounter, Order, Result and Referral with consent attached, so authorised views are consistent.
- Clinicians decide what is recorded. Privacy officers set access rules.
- Logistics
- Shipments, containers, vehicles, drivers and documents are tracked in different systems across a corridor.
- The ontology links Shipment, Container, Vehicle, Document and Milestone, so delay and cost questions have one answer.
- Operations managers decide on rerouting and exceptions.
### Built on
**Zeus**
- Is the home of the ontology: studio, versioning, connectors, query layer and APIs for people and agents.
- Boltcliq product
- One source of business meaning.
**Odin**
- Decides who and what may read objects and run actions, using attribute and relationship based rules.
- Boltcliq product
- Permissions follow the model.
**Monolith**
- Governs which parts of the ontology an agent may load as context and which actions it may call.
- Boltcliq product
- Agents see only what the task needs.
### What we do and do not do
- Model with your domain experts in workshops
- Map objects to your existing systems
- Define governed actions with permissions
- Set up model governance and ownership
- Train your team to maintain the model
- Model the whole enterprise at once
- Impose a generic template without testing
- Hide the model inside a black box
- Let agents write directly to source tables
- Replace the judgement of domain owners
### Questions we hear
- What is an ontology, in one sentence?
- It is a shared, versioned description of the objects in your business, how they link and what can be done with them, used by people, applications and agents.
- How is it different from a data model or a data warehouse?
- A warehouse stores data. An ontology defines meaning, relationships, rules and actions, and sits over the data. It also supports actions that change the world, not only queries.
- Do we need Zeus to have an ontology?
- No. The method works on other tools. Zeus is built for the job and adds studio, governance and agent interfaces.
- How long does a first ontology take?
- A first domain can be modelled and tested within weeks. Time depends on the number of systems, the quality of data and availability of domain experts.
- Who owns the ontology?
- You do. It is your description of your business. We document it in open formats and hand over the governance process.