← Back to the platform

The ARPIA ontology

A shared layer where people, AI agents and apps work with your organization’s data and systems, safely. It turns every database and every MCP system into business objects with rules, permissions and a history.

Connecting AI to a company’s systems usually means custom work for each one: a connector, prompts that explain what the data means, a set of permissions, and then keeping all of it current as the systems change. Every new agent repeats the job, and every one ends up with its own version of the truth.

The ontology does that work once, in one place. Your teams define what a customer, an order or a risk means for your organization. From then on, every agent, app and person works through the same definitions and the same rules, and everything they do is recorded.

WATCH THE OVERVIEW from your systems to governed business objects

What changes with the ontology.

Without an ontology

Agent AAgent BApp
Agent A → ERP (42 tools) · CRM (35 tools) · Tasks (28 tools)
Agent B → ERP (42 tools) · CRM (35 tools)
App → database (direct access)
  • Each connection brings its own tools, with no business context
  • Permissions are set connection by connection
  • Every agent interprets the data its own way
  • Nothing ties together who did what

With the ARPIA ontology

Agent AAgent BApp
↓
ONTOLOGY
CustomersOrdersInvoicesTasks
definitions · permissions · approvals · log
↓
ERPCRMTasksDatabases
  • Agents see business objects, not hundreds of tools
  • One set of permissions and approvals for everyone
  • One definition of each concept, with an owner
  • Every read and every action is recorded
Before and after the ontology: the same systems, reached through one governed layer.

How the ontology works.

The ontology is made of three pieces.

Nodes
A business entity, such as a customer or an order. A node can come from a table, a query or an MCP system, and it can relate to nodes from any other source.
Relations
How nodes connect: a customer has orders, an order has invoices, even when each one lives in a different system.
Actions
What can be done to a node, such as approving a discount or creating a task. Each action declares who may run it and whether it needs approval.

Reading and writing work the same way for every source

An agent does not need to know where the data lives or which tool to call. It asks the ontology, in business terms, and the ontology applies your rules on the way in and on the way out.

read orders where status = "at risk" this week
→ 18 rows, joined with the CRM account and its owner

write discount.approve(order 4815, 8%)
→ waiting for approval by a sales manager

The rules themselves are written by your teams in plain language: what an active customer is, when an order is at risk, what margin really means. The ontology keeps each one as a definition with an owner and a change history.

Three ways to use it

ORIA

Ask in conversation. ORIA answers and acts through the ontology, with the permissions of the person asking.

Your apps and flows

Apps, dashboards and scheduled flows read and act through the same nodes and actions.

Outside agents, through MCP

Agents you already use, such as Claude, connect to the ontology over MCP and get the same rules.

An example

A sales director asks ORIA which large customers have orders at risk this month.

The question crosses three systems: orders live in the ERP, accounts in the CRM, open tasks in the project tool. ORIA reads the three nodes through the ontology, joins them by customer, and answers with a table.

Then it proposes a follow-up task for each account owner. Creating tasks is an action that needs approval, so the director reviews the list and approves it. Every read and every task is recorded.

Finally, the director saves the whole thing as a weekly flow. It runs with the same definitions and the same permissions every time.

Every MCP, inside the ontology.

Most of your systems already speak MCP. Connected directly to an agent, each one arrives as a long list of tools. ARPIA brings them into the ontology instead, next to your databases.

Organized by business entity

Each MCP server joins as nodes and actions grouped by what they are about: customers, orders, tasks.

Fewer tokens

The agent loads the business objects its task needs, not every tool of every server. Data is filtered and calculated before it reaches the model.

Governance

Reading needs a permission. Writing needs a permission and an approval. Sensitive columns are masked by profile, and credentials never leave the platform.

Observability and change

Every read and every action is logged. When a system adds or changes tools, the ontology notices and the change waits for review or applies with an undo.

Where the ontology is already at work.

CIEPS

Disinformation Observatory

With CIEPS-AIP, Panama

An observatory that studies disinformation in Panama’s digital environment with science and evidence, running on ARPIA.

More in ARPIA Labs
InterFuerza

AI in business operations

With InterFuerza, Grupo ENX

Businesses on InterFuerza connect sales, inventory, collections and accounting to AI through MCP, with governance through Oria by ARPIA.

More in ARPIA Labs

What the ontology does not do yet.

  • It does not decide your definitions. Someone in your organization has to own each rule, and the first ones take real work.
  • Systems without a database connection or an MCP server need a connector before they can join.
  • Agents still make mistakes. That is why writing goes through actions with permissions and approvals, and why everything is recorded.
  • New fronts on the same ontology, like equipment and sensors, are being explored in ARPIA Labs.

What does your organization know that your AI doesn't?

Bring one business definition your teams argue about. We will show you how it becomes a governed, queryable rule.

Talk to an Expert Back to the story