DataFocus · The enterprise foundation for AI

Your business context. Your rules. Any agent.

Connect business databases and documents, organize knowledge with a business ontology, and serve governed context through open interfaces such as MCP, so every use of enterprise knowledge is grounded, bounded and traceable.

Two editions: OT Flash and OT Heavy
S1Order delivery riskAsked by EAST sales leadSnapshot 2026-09-25 10:00
Q

Which EAST orders due September 28–30 might be late? Show me the evidence. Don’t change orders or send messages.

A
  1. O-1001Forecast 2 days late

    Under this company’s , that is a predicted risk, not a breach.

  2. O-1002No sign of delay in current data

    That doesn’t guarantee it will be on time.

  3. O-1004Not enough information

    Production has no completion forecast yet. It can’t be assessed, and it shouldn’t be counted as on track.

NextAsk production planning for an O-1004 forecast. A follow-up note for O-1001 is drafted, not sent.

Visible scope: EAST. Identity sets the scope, not the wording of the question.0 writes to source dataSynthetic data

S1 · Enterprise task

One task needs both data and business knowledge.

Orders live in ERP, production progress in MES, and what counts as delivery is written in the contract. Only together do they show the risk, and only after filtering by the asker’s identity can the result go to this sales lead.

OrderDueForecastFindingEvidence
O-100109-2809-30Forecast 2 days lateTwo days late by Contract A v2’s definition of delivery. A forecast, not a breach; no penalty is computed.ERPMESContract A v2
O-100209-2909-29No sign of delayNo sign of delay in this snapshot. That’s not a promise it will be on time.ERPMESContract A v2
O-100409-30No recordNot enough informationNo completion forecast. Needs follow-up and can’t be marked on track.ERPMESContract C v1
Synthetic data from a demo enterprise, snapshot 2026-09-25 10:00 (UTC+8)

What if the sales lead asks for O-1003 directly?

QWill O-1003 be delivered on time?

Check each source in the walkthrough

Why a database connection isn’t enough

You have agents. The context should still be yours.

When a company uses several agents at once, who manages the business knowledge they receive? Give each tool its own copy, and definitions and permissions soon drift apart. General-purpose agents are capable, but they can’t answer the questions below on their own.

  1. Entities

    Is “Customer A” in CRM, ERP and the contract the same company?

  2. Definitions

    Does delivery mean factory completion, dispatch or customer sign-off?

  3. Valid knowledge

    The contract was revised. Which version applies now?

  4. Visibility

    Can the EAST sales lead see WEST orders?

Connected doesn’t mean understood. Queryable doesn’t mean permitted.

Owning your context is about more than where the servers sit. It means your enterprise can:

  • DefineOwners confirm objects and definitions
  • ApproveChanges take effect only after approval
  • MaintainKnowledge is updated when rules change
  • AuthorizeThe enterprise decides who reads and who edits
  • VerifyEvery answer traces to a source and version
  • MigrateDefinitions, mappings and policies should be exportable

Understand · Business ontology

Every business term, defined your way.

DataFocus organizes objects, relationships and definitions as a business ontology: customers own orders, and orders link to production tasks and the contract in effect. Every field has a source and every definition an owner, so agents look up a definition instead of guessing from table names.

Select a term to read its definition

Customer owns Order; Order fulfilled by Production task; Order governed by Contract in effect

Delivery

ContractDoc · deliveryBusiness definition

In this demo enterprise, “delivery” means production is complete and ready for handover at the factory. Transport time and customer acceptance are excluded.

So “forecast two days late” is a prediction under this definition. It is not a breach and triggers no penalty.

A real company’s delivery definition must be confirmed by its own business owner.

From connection to trail

  1. ConnectBring business data and business knowledge into the same task.
  2. UnderstandOrganize information by the company’s definitions, so agents don’t guess at tables.
  3. AuthorizeShared knowledge doesn’t mean shared visibility.
  4. UseOffer capabilities with business meaning, never a raw pipe to the whole database.
  5. TraceKnow how knowledge was used, and how it changed.

Authorize · Trace · S3

Versioned knowledge. Bounded access. Recorded change.

The support policy tightens from a 24-hour to a 4-hour response, effective from midnight on September 25. Until the owner approves it, agents keep answering with the old rule. After approval, the same ticket’s deadline moves 20 hours earlier.

T-2001 · premiumSynthetic data
  1. 09-25 00:00POLICY-v2 becomes business-effective
  2. 09-25 09:00Ticket T-2001 created
  3. 09-25 10:00Demo snapshot; policy approved at this point
  4. 09-25 13:00v2 deadline (4 hours)
  5. 09-26 09:00v1 deadline (24 hours)
The draft says four hours, but it isn’t approved. Agents still give September 26, 09:00 and cite v1.
  • Bounded accessSame knowledge, different scope per identity. Out-of-scope requests are refused without revealing whether the target exists.
  • Approved changeDrafts never change answers. Approval binds a specific version and scope; if the version changes, it must be reviewed again.
  • A usable trailEvery call records sources, definitions and returned scope. Every change records before and after, the diff and who approved it.

Use · Many entry points

Change the agent. Keep the knowledge.

The agents people use can change with the team. Definitions and access rules stay in the enterprise layer, and every entry point calls them with the same meaning and permissions. Wording can differ; facts and scope should not.

Entry point A

O-1001 is forecast two days late, judged by the factory-handover definition. O-1004 has no completion forecast, so it can’t be judged yet.

Entry point B

One order may slip: O-1001, by about two days, defined as factory-ready handover. O-1004 has no production forecast and needs more information.

Synthetic-data example: two entry points use different wording with the same definitions, sources and scope. Each workspace manages its own conversations.

Choose your workspace

Plan access around the client version, operating mode and enterprise permissions. These names help you choose and evaluate; they do not imply an official partnership.

See connection requirements

Build · S2

Start with one task. Build it with people and agents.

Coding agents such as Codex can inspect sources, propose objects and relationships, and generate diffs and tests. When a definition is unclear, the agent should stop and ask. Before anything is published, someone with authority approves it.

Explore the building workflow with synthetic data. For your own instance, use the tools and permissions available in that version.

  1. Coding Agent
    Inspect sources

    CRM customers · ERP orders · MES production tasks · contract documents

  2. Coding Agent
    Propose a draft

    4 object types · 3 relationships · 1 definition to confirm

  3. Coding Agent
    Stop and ask

    Does “delivery complete” mean production finished, goods shipped or customer acceptance? It changes the risk result, so I won’t decide it for you.

  4. Coding Agent
    Diff and tests

    Source writes 0 · access unchanged

    + Customer → Order
    + Order → ProductionTask
    + Order → ContractDoc
  5. Owner
    Wait for approval

    Publishing needs someone with authority. The agent can’t approve its own change.

  6. Business agent
    Hand to a business agent

    Once published, another agent runs S1 on the same definitions.

S4After the first project, definitions, mappings and tests move to the next client. Customer data and access grants stay behind.See how a delivery is reused

Adopt · Two editions

Run it in your environment. Start with one business domain.

DataFocus OT comes in two editions, both with ontology management. OT Flash uses Apache Doris as its in-memory compute engine and its data warehouse: lighter to deploy, fast to query. OT Heavy continues the Hadoop, Hive, Trino and Iceberg architecture and adds ETL and AutoML. With either edition, start with one task that has data, documents, a business owner and a result someone can accept.

OT = OntologyOntology managementIn both editions

DataFocus OT Flash

One engine. A fast start.

Apache DorisIn-memory compute · data warehouse

Apache Doris serves as both the in-memory compute engine and the data warehouse.

  • Ontology management
  • Query and analyze Apache Doris brings analytical queries and the data warehouse into one engine.

Suited forTeams that want one business domain running quickly, with data that can sit in a single analytical engine.

Free distribution View releases and downloads

DataFocus OT Heavy

A full big-data foundation.

TrinoDistributed SQL queries
Apache HiveData warehouse
Apache IcebergOpen table format
Apache HadoopDistributed storage and compute

Continues the Hadoop, Hive, Trino and Iceberg architecture, with ETL and AutoML built in.

  • Ontology management
  • ETL
  • AutoML
  • ETL Extract, transform and load data.
  • AutoML Automated machine-learning modelling.

Suited forEnterprises with large data volumes, an existing big-data platform, and a need for data pipelines and modelling.

Free distribution View releases and downloads

OT stands for Ontology. Choose an edition around your data scale, existing stack and implementation needs.

Core
Closed source
Adoption
Free distribution and local deployment for both OT Flash and OT Heavy
Distribution, commercial and support terms
Use is governed by the license and service terms for the selected edition
Installers
Free installation images on GitHub, with release and verification information
Running costs
You pay for servers, model calls and third-party services
Delivery
Self-host free, work with a builder, or discuss a hosted deployment

Your enterprise owns the knowledge and rules. Different agents do the work.

Start with the demo enterprise. No customer data required.