Connected data
Documents, master data and project information retain links to their sources. Where a task needs clean reference data, MDM supports matching, golden records and a history of decisions.
Enterprise AI platform
AI for your corporate data and workflows
Nexus brings corporate data, tools and business workflows together in one AI platform. Your team approves the operating rules and resolves uncertain cases; the platform records sources, actions and decisions.
- Runs in your infrastructure
- You approve model providers
- Rules approved by your team
- Sources and audit trail
Why this exists
What an AI pilot needs before it can become daily work
A useful answer in a demo is only a start. In daily work, an agent needs access to the right sources, permission to use tools and a clear route to a person when the evidence is incomplete or conflicting.
Nexus brings these parts together: connected data, approved workflows, access checks and an audit trail. The first task determines which sources and components you need.
The market reality
Data readiness and the growth of enterprise AI
60%of AI projects unsupported by AI-ready data will be abandoned through 2026Gartner forecast, Feb 2025
7%of respondents say their organization's data is completely ready for AICloudera × HBR Analytic Services, Mar 2026
40%of enterprise apps will feature task-specific AI agents by end-2026 — up from under 5% in 2025Gartner forecast, Aug 2025
Where to start
Choose a task. Prepare the data it needs.
Start with document search, project checks, master-data cleanup or another defined task. Review the sources it needs and agree how people will check the results. MDM is one option; it is not a prerequisite for using Nexus.
Choose a taskDefine the result, sources and responsible people
Agree the rulesSet permissions, checks and escalation points
Check the resultRun the scenario and review it with your team
The scope starts with your task
What makes it work
How the platform works
Agents and approved workflows
People approve the rules and permitted tools. Agents work within those limits; disputed cases and decisions requiring separate approval go to a responsible person.
Deployment and model choice
Nexus runs in your infrastructure and can send requests to public AI services you approve. On-premises models are an option when their quality meets the task requirements.
Access and audit
Agents act within the rights of the user who launched them. Source references, tool calls and recorded decisions let your team check how a result was produced.
The technical core
How Nexus gathers and checks information
ARBA combines context management, parallel agents, an organizational graph and verification methods in a shared platform core.
Context across sources
The system connects information across available documents and records, so a task can use related sources together.
Parallel execution
A complex task is split between agents. Their results are brought together and checked before the final response.
Organizational graph
Entities, roles, ownership and hierarchies connect knowledge to the structure of the company and the people responsible for it.
Cross-checking results
Agents check proposed conclusions against sources and each other. Tool calls and intermediate results are recorded for review; unresolved questions need human judgement.
Bitemporal history
The data model records when information applied and when the system learned it — “as-was / as-known”. This supports checks against the historical state of the data.
Hybrid methods
Deterministic rules, semantic search and probabilistic methods serve different parts of a task. The workflow combines them with checks and human decisions.
Our own practice
AI in daily work at ARBA
Our team uses AI agents in development, support and coordination. This work helps us refine operating rules, review procedures and the handling of exceptions.
The platform, module by module
A shared core for four components
Modules share identity, access checks, audit and cost accounting. Your team approves the rules and handles disputed cases. Choose the components your task needs.
Nexus AI Foundation
The agentic AI platform every module is built on
Agents & scenarios
- AI agents with tools and memory
- Configurable task scenarios
- Orchestration of multi-step tasks
- Recorded tool calls, results and decisions
Landscape integration
- Single sign-on via your corporate directory
- Approved model services, including Microsoft Foundry
- Agents and a bot inside Microsoft Teams
- Connectors configured for your sources
Security & privacy
- Roles, permissions and a security-officer role
- A full audit log of every action
- Outbound-transfer (egress) control
- PII de-identification; instance isolation
Models, languages, cost
- A per-purpose model selector
- Multilingual — Russian, English, Kazakh
- Model usage and cost monitoring
- The platform meters neither tokens nor records
Master data, in depth
When one item has five codes
An item may appear under several codes, with key differences buried in descriptions and stock split across records. Nexus helps reconcile these records while preserving sources, approved rules and links to every legacy code.
The master-data presentation explains the workflow, project deliverables and responsibilities, from the first extract to prepared data for your target systems.
The architecture
Three data planes and one capability plane
Nexus has three data planes — Master Data, Documents and Project Control — and a capability plane for scenarios, tools and connectors. Use the components your task needs; a full MDM rollout is not required.
- Master Data — matching, rules for choosing golden-record values, reversible merge and bitemporal history
- Documents — retrieval within user permissions, typed metadata, OCR in Russian, Kazakh and English, and enrichment
- Project Control — checks reported progress against project documents and flags gaps for review
- Capability plane — scenarios, tools and connectors to existing systems
The stack
Data, tools and agents in one architecture
Your AI — agents and copilots
Capability plane
- Scenarios
- Tools
- Connectors
Master DataGolden records, matching, survivorship
DocumentsPermission-aware retrieval, OCR, enrichment
Project ControlEvidence-grounded assurance
Platform in your infrastructure · Approved model providers
Integration with your systems
Connect Nexus to the systems you already use
Nexus uses data and tools from your Microsoft environment and enterprise systems. The required connectors, permissions and direction of data exchange are agreed for each project.
Your infrastructure and connected services
Nexusa single agentic AI platform
Microsoft ecosystem
- Active Directory · SSO
- Microsoft Foundry · models
- Fabric & Power BI · data & BI
- SharePoint · documents
- Teams · collaboration
Enterprise & industry systems
- SAP
- Oracle
- 1C
- Schedulers · Primavera P6
- Integration bus · ESB
- Data warehouse · BI
The diagram shows integration areas. Connector availability and setup are confirmed for your project. Model requests may go to an approved public AI service.
Applied solutions
Tasks you can start with
Each scenario uses the platform components and integrations it needs. Select a task to see the problem, the workflow and the expected result.
Master-data deduplication & golden records
CDO / Head of Data- The problem
- Duplicates and conflicting reference data make stock checks, purchasing and reporting harder.
- What Nexus does
- Matches counterparties and materials, proposes golden records and preserves source codes. Data owners approve rules for routine cases and review disputed matches.
- The outcome
- Prepared records, a source-code mapping and a history of decisions for transfer to your systems.
Your team can configure further scenarios. Their rules, tools and permissions are reviewed before use.
For your team
What each role needs to know
Select a role.
The task
Connect AI to existing systems while keeping architecture, access and costs under control.
How Nexus helps
A shared platform for agents and integrations, with agreed providers, permissions and operating rules.
For the CIO
Deployment, data processing and human control
- Deployment — Nexus runs in your infrastructure. Calls to public AI services send the request content to the approved provider; hosting the platform yourself does not make inference local
- Model choice — public services, including OpenAI and Anthropic Claude, or on-premises models whose quality is sufficient for the task. Providers, regions and data-processing terms are agreed before use
- Outbound access — model endpoints and permitted data flows are configured explicitly. Unapproved endpoints are blocked
- User permissions — agents act on behalf of the user who launched the task; authorization is checked for each request
- Human oversight — responsible people approve operating rules and permitted tools. Disputed cases and actions requiring separate approval return to a person
- Audit and protection — recorded tool calls and decisions, document access logs, separation of duties and encryption of source data and derived information
- Isolated deployment — requires local models and dependencies, an agreed offline update process and validation of the selected tasks. Public AI services require connectivity
How Nexus compares
How to assess Nexus alongside other approaches
Compare the scope of the task, existing systems, integration work and operating model. Capabilities vary by product and configuration.
What to assess
Model access, tools, hosting options and integration with your existing cloud services.
The Nexus approach
Runs in your infrastructure and connects to approved public AI services or suitable local models. Model hosting and data flows are agreed separately.
What to assess
How the platform represents business entities and processes, and what it takes to connect and maintain them.
The Nexus approach
Combines an organizational graph with documents, master data and project information. Start with the components needed for one task.
What to assess
Long-term data governance, stewardship workflows, domain coverage and ERP integration.
The Nexus approach
Prepares and reconciles reference data, preserves source codes and decisions, and connects this work to other AI scenarios. It can prepare data for an existing MDM system.
What to assess
The engineering needed for access control, integrations, approvals, audit and ongoing support.
The Nexus approach
Provides shared platform mechanisms for those tasks. Your team configures workflows within agreed permissions and rules.
This is a guide to comparing approaches. Specific products may offer overlapping capabilities; assess them against your requirements.
How adoption works
Start with one task and expand after review
- Choose a taskAgree the result, sources and responsible people. Documents, project checks and master data are possible starting points.
- DeploySet up the platform, access rights, integrations and approved model providers.
- ReviewApprove the rules and test the workflow. People resolve disputed cases and assess the result.
- ExpandUse the results to select further tasks and the components they need.
- DevelopYour team configures new scenarios and tools, with review before activation.
Delivery network
Regional implementation and support
Regional partners help with implementation, integrations and support. ARBA is responsible for platform engineering. The delivery team and responsibilities are agreed for your project.
Impace GroupEurope · Middle East · Central AsiaA regional network of transformation companies with SAP-centred enterprise implementation at its core — business consulting, system implementation and infrastructure delivery.
TeleconKazakhstanAn enterprise implementation and consulting company with 100+ specialists and 18+ years of SAP partnership — deep local delivery capacity and knowledge of Kazakhstan's regulatory environment.Straight answers
Scope and responsibilities
- Nexus works alongside SAP, Oracle, 1C and Microsoft. The integrations and permitted actions are defined for each project
- People approve the operating rules and resolve disputed cases. Preparing master data and loading it into an ERP are separate responsibilities
- Master-data matching is available for the first domains; volume and quality requirements are checked on your data before acceptance
- Nexus is a platform for using models. On-premises models must meet the quality requirements of the selected tasks
Practical questions
Nexus is deployed in your infrastructure. When a workflow calls a public AI service, the request content is processed by that provider. Endpoints, regions and permitted data flows are agreed in advance. Fully local processing requires suitable local models and dependencies.
Nexus can use public AI services, including OpenAI and Anthropic Claude. On-premises models are also an option if their quality is sufficient for your tasks. We assess the choice against quality, cost and data-processing requirements.
Responsible people approve operating rules and permitted tools. The system applies those rules; disputed cases and decisions requiring separate approval return to a person. Your team can describe new workflows, which are reviewed before activation and remain bounded by user permissions.
No. Start with the task you need: for example, document search or a project check. We assess the sources and preparation needed for that task. MDM is useful when the task requires reconciliation and cleanup of reference data.
ARBA develops the platform and works with regional partners, including Impace Group and Telecon. The project defines who handles deployment, integrations, training and support.
Yes, with local models whose quality meets the task requirements. An isolated deployment also needs local dependencies and an agreed offline update process. We validate the selected tasks in that configuration before acceptance.
We agree a defined task, required sources and acceptance criteria. The first project checks the result, quality and operating costs on your data. Timing depends on data availability, integrations and participation from your team.
Discuss your first task with Nexus
Tell us what your team needs to do. We will discuss sources, integrations, model providers and how to assess the result.
Executive briefing
Book a Nexus briefing
Describe the task and the systems involved so we can prepare for the discussion.