We create a single source of truth and depth for your company analysis.
Agentica Onta is a digital model of your company established on existing systems: your data, logic, rules, metrics. It is enabling people, applications and AI agents to answer, reason and act based on a single source of truth.
The problem worth naming
Somebody was asked that question this quarter. Somebody is probably still working on the answer.
01Why it's hard
It isn't a hard question. It's hard because of how the business grew. Every market you entered arrived with its own systems. Every channel you added counted things its own way. Every acquisition brought a platform nobody ever retired. You now sell hundreds of products across all of it, and the record of that sits in a dozen systems that were never built to be read together.
02The manual fix
So someone pulls the exports, lines up the product codes by hand and reconciles the numbers in a spreadsheet. It works, once. By the time the answer lands the quarter has moved on, and the next question starts the whole assembly again.
03Add an agent
Pointing an AI agent at the same systems doesn't change the arithmetic. It has to work out on its own what each system means by "product" or "revenue" — and the next agent starts from zero again.
04What actually fixes it
One model of the business, on the systems you already run. Products, customers, revenue and the rules between them are defined once and point at live data, so every person, report and agent reads the same answer — with its derivation attached.
What it is
An unified data and logic layer of your business.
Agentic Onta sits between your systems and everyone who needs an answer from them.
- People
- Agents
- Applications
- BI
Agentic Onta
- CRM
- ERP
- Warehouse
- Document
- Operations
No copying
It maps onto your existing source
No ingestion
Your data stays where it is
One definition
Change it once, everywhere
How it works
Four steps. The first three take days.
01
Connect
Point Agentica Onta at your sources. Read-only, through the credentials and permissions you already have. You build no pipelines.
Name Entities Examples CRM Account · Opportunity · Activity Salesforce · HubSpot · Dynamics ERP Invoice · Material · Cost Centre SAP · NetSuite · Infor Warehouse Segment · Price List · Period Snowflake · BigQuery · Databricks Documents Contract · Specification · Policy Google Drive · OneDrive · NextCloud Operations Work Order · Incident · Shift ServiceNow · Jira · Workday Shop floor Batch · Machine · Downtime Siemens Opcenter · AVEVA · Ignition Support Case · SLA · Sentiment Zendesk · Intercom · Freshdesk 02
Model
Visual editor or code. Most teams have a working first model inside a day.
Build the ontology whichever way suits your team:
- Bottom-up
- Generate a draft from your existing schemas, then correct it.
- Top-down
- Start from SQL DDL, or describe the domain and let an LLM draft it.
- Import
- Bring an ERD or an OWL ontology you already have.
- Catalogue
- Turn your data catalogue into a live model.
03
Map
Connect the model to the real columns underneath. It's queryable immediately. No load step, no refresh window, no second copy of the truth to keep in sync.
- Revenue
orders.revenue_amount - Customer
crm.customer_id - Product
product.product_id
- Revenue
04
Serve
Publish metrics and data products to a library any tool can reference. Switch on the agent endpoint. Your BI, your apps and your agents now share one source of truth.
Name Tools Ask Chat · Explorer · Slack & Teams Automate MCP server · A2A · Event triggers Build REST · GraphQL · Python SDK Analyse Power BI · Tableau · Iceberg tables
Time to value
What the first ninety days actually look like
Onta is a platform, not a plug-in. Modelling a business that runs across several markets, multiple channels and hundreds of products is real work, and anyone who tells you otherwise is selling you a demo.
01Week 0 — Vision and discovery
Map systems, owners, business questions and candidate use cases, then put a value against each one.
You get:
- Data estate map
- Prioritised use cases
- Value proposition
02Weeks 1–2 — Technical scope
Examine connectors, data quality, access and join risks.
You get:
- Technical scope
- Awkward sources
- Join risks
03Weeks 2–6 — Connect and model
Connect existing sources and build the first ontology — entities, relationships and rules.
You get:
- Cross-system answers
- Team modelling unaided
04Weeks 6–12 — Production
Publish metrics, permissions, provenance and the agent endpoint. Put the first use case in front of real users.
You get:
- First data product in production
- Manual quarterly assembly retired
05Month 3 onward — Expand
Add more domains, sources and agents on the existing model.
You get:
- Second use case in weeks, not months
Capabilities
What you actually get
Business people can read the model
Model entities, relationships and rules in a visual editor or in code. Version them, review them, promote them. A model only your data engineers can read isn't a shared understanding — it's another silo.
Numbers that are never stale
Every class and property points at a real column in a real system. Query the ontology and the query hits live source data. Nothing goes out of date, because nothing is a copy.
One definition of revenue, everywhere
Define a KPI once and reference it from BI, from an API call, from an agent. When finance changes the definition, every consumer changes with it. Nobody reconciles anything.
Agents that reason on your business
Agents connect over MCP and work against the model. They inherit your permissions, so they can't surface anything the person asking couldn't already see.
Every answer shows its working
Results trace back to source, logic and filters. Someone who wasn't in the room can check the number six months later — which is the only kind of answer that survives an audit.
Runs where your rules require
Your infrastructure, your private cloud, EU-resident cloud, or fully disconnected. Your data doesn't leave your perimeter to be reasoned over.
Use cases
What we can build on it
Supply chain
See a shortage before it lands. Suppliers, purchase orders, stock and demand in one model, so a late shipment shows you every order and customer it touches.
Finance & controlling
One definition of revenue. The number in the board pack matches the dashboard matches the number the agent just gave your VP, because all three read the same model.
Customer & revenue
Find the revenue sitting between your systems. Renewal risk and cross-sell exposure live in the relationships between CRM, contracts and support.
Quality & operations
Trace a defect to its cause, not to a system boundary. Engineering, production and after-sales in one model, so root-cause analysis follows the relationships all the way down.
Regulated reporting
Report from the model, not from a spreadsheet. Every figure carries its lineage back to source, so an auditor can follow a number without asking anyone to rebuild it.
How this differs
Three things get called “semantic.” They do different jobs.
| The question it answers | Where it stops | |
|---|---|---|
| BI semantic layer | Do all our tools calculate this metric the same way? | It defines queries, not concepts. What "revenue" means still has to be read back out of the SQL by whoever arrives next. |
| Property graph | How is everything connected, so I can traverse it? | It records how your facts connect, not how your concepts connect. Meaning still has to be rebuilt from labels and edges. |
| Agentica Onta | What do these things mean, and which rules apply when a machine acts on them? | Meaning sits in the model next to the data. An agent reads it instead of inferring it. |
Security and governance
Nobody sees what they shouldn't. Agents included.
- Permissions carry over. People and agents see what the source systems already allow them to see. You don't maintain a second permission model.
- Answers stay traceable. Source, logic, filters — reviewable by someone who wasn't there.
- Your identity provider. SAML, OIDC, OAuth 2.0, through the IdP you already run.
- Your jurisdiction. Your infrastructure, your private cloud, EU-resident cloud, or disconnected.
- Your model. Bring your own LLM or run an open one inside the perimeter. Swapping the model doesn't touch the meaning.
FAQ
What people ask before they commit
Do we have to move or migrate our data?
No. Agentica Onta reads your sources where they are — read-only, through the credentials and permissions you already have. Every class and property in the model points at a real column in a real system, so there is no load step, no refresh window and no second copy of the truth to keep in sync. You build no pipelines.
How long until we're in production?
Most teams have a working first model inside a day. The first data product is in production within twelve weeks: discovery in week 0, technical scope in weeks 1–2, connecting and modelling in weeks 2–6, production in weeks 6–12. After that, a second use case takes weeks, not months, because it builds on the model you already have.
We already have a warehouse and a semantic layer. Does this replace them?
No — it sits on top of them. Your warehouse stays a source Onta reads from, and your BI keeps working: it references metrics published from the model instead of redefining them. What changes is where meaning lives. A semantic layer defines queries; Onta defines what things are and which rules apply, next to the data.
How do permissions work?
They carry over. People and agents see what the source systems already allow them to see, so you don't maintain a second permission model. Sign-in goes through the identity provider you already run — SAML, OIDC or OAuth 2.0 — and an agent can never surface anything the person asking couldn't already see.
How is this different from RAG or a vector database?
RAG retrieves passages that look similar to the question and leaves the model to infer what they mean. Onta gives agents the meaning itself: entities, relationships and rules, each pointing at live data. The agent reads the definition instead of guessing it, and the answer comes back with its derivation.
What stops the AI from hallucinating?
Agents answer against the model, not from memory. Every result traces back to its source, logic and filters, so someone who wasn't in the room can check the number six months later. If an answer can't be derived from the model, it doesn't get made up to fill the gap.
Can agents act, or only read?
Both, within limits you set. Onta's own connections to your systems are read-only; agents act through the tools you switch on — the MCP server, A2A and event triggers — under the rules the model defines and the permissions of the person they act for.
How is it licensed?
As a platform for your organisation, not per seat or per query. The price depends on the domains, sources and deployment you need — the working session is where we scope it with you.
Where does it run?
Where your rules require: your infrastructure, your private cloud, an EU-resident cloud, or fully disconnected. Your data doesn't leave your perimeter to be reasoned over. Bring your own LLM or run an open one inside the perimeter — swapping the model doesn't touch the meaning.
Every new agent starts from zero. Unless they share a model.
Bring one messy domain and two source systems. In 30 minutes we'll model it and show you answers coming back with their derivation attached — on your data, not a sandbox. If we're wrong, you'll know inside the half hour. That's rather the point of doing it on your data.