NEXEME ATLAS

Your SAP system isn't standard SAP.

Modern AI models know a surprising amount about SAP.

But they don't know the SAP system your organization has built over years or decades.

They don't know your Z-tables and custom fields.

They don't know your configuration and coded values.

They don't know which objects matter in your landscape or how they relate to one another.

And they don't necessarily know what your business users mean when they use apparently simple terms such as customer, revenue, sales, margin or available stock.

That matters when AI moves from answering questions about SAP to answering questions using your SAP data.

Giving an AI agent access to SAP isn't the same as giving it an understanding of your SAP.

Imagine a business user asks:

“List my top 3 customers by revenue with their city and country.”

Simple question.

But before an AI agent can generate a reliable query, it needs to answer questions the business user never asked explicitly.

What does “revenue” mean in this organisation?

Is revenue represented by SD billing, FI postings or another source?

Which document types count?

Are cancellations and credits included?

Which amount and currency fields are appropriate?

What does “customer” mean?

Sold-to party?

Bill-to party?

Payer?

Another customer relationship defined by the organisation?

Where do city and country come from?

Which customer or address objects contain them?

Are standard SAP objects used, or has the organisation customised them?

And how is everything connected?

Which tables, fields and relationships actually exist in this SAP landscape?

This is the context an AI model cannot safely infer from generic SAP knowledge alone.

Business User

“List my top 3 customers by revenue with their city and country.”

↓

AI Agent

Understands the user's intent and identifies the SAP context it needs.

↓

NEXEME ATLAS

Meaning + verified SAP structure

Atlas helps the agent establish:

What does “revenue” mean here?

What does “customer” mean here?

Which tables and fields actually exist?

Are custom objects involved?

What coded/configuration values matter?

How are the required objects related?

What are the valid join paths?

↓

AI Agent

Builds the query plan and generates SQL

Using the context supplied by Atlas, the AI determines the required objects, joins, filters, aggregation and ordering and generates the appropriate query.

The proposed query can then be checked against Atlas:

Do the tables exist?

Do the fields exist?

Are the proposed relationships supported?

Has the agent invented anything that isn't present in the customer's landscape?

↓

Authorized Data Tool

Executes under the customer's existing controls

The SQL is executed by an independently authorized data-access component against the approved business-data source.

Atlas does not contain the underlying customer transactions.

The execution layer remains responsible for access control, permitted data, read-only enforcement and other enterprise security policies.

↓

AI Agent

Receives the authorized result and answers:

1. Customer A — $24.7m — Singapore, Singapore
2. Customer B — $18.2m — Sydney, Australia
3. Customer C — $15.9m — Tokyo, Japan

Illustrative result only.

YOUR AUTHORIZED SYSTEMS PROVIDE THE DATA.

This separation is deliberate.

Atlas doesn't replace your SAP system.

It doesn't replace your AI platform.

And it doesn't need to become another business-intelligence tool.

NEXEME Atlas gives AI the customer-specific SAP context it needs to reason more reliably about your enterprise data.

Your organization remains in control of which AI applications can access which business data and what they are permitted to do with it.

A context layer between your SAP landscape and your AI agents.

Atlas reads your SAP system's own data dictionary and repository metadata and compiles it into a self-contained Knowledge Artifact.

The artifact maps the structure of your actual SAP landscape, including:

Tables and fields
Including custom objects and technical definitions.

Relationships and join paths
Derived from relationships in your own SAP system.

Configuration and coded values
Helping agents interpret values instead of guessing what they represent.

Customisation
Making customer-specific objects visible alongside standard SAP.

Landscape prominence
Helping agents focus on structurally important objects rather than treating hundreds of thousands of SAP tables equally.

AI agents can query this knowledge through a controlled interface whenever they need SAP context.

1 — BUILD

Atlas connects to SAP using controlled, read-only access to the data dictionary and repository catalogue. Or required data is replicated to customer’s data lake. Or customer provides data as JSON files. 

It discovers the customer's actual SAP landscape and compiles the results into a schema-versioned Knowledge Artifact.

↓

2 — UNDERSTAND

AI agents query Atlas when they need to understand the SAP landscape:

Which objects are relevant?

Which fields exist?

What do coded values represent?

How are these objects connected?

Is there a custom object I need to consider?

↓

3 — REASON

The AI combines the verified structural context from Atlas with business meaning and customer-specific semantics to construct an appropriate query plan.

↓

4 — VALIDATE

Tables, fields, relationships and other structural assumptions can be checked against Atlas before a generated query reaches the business-data layer.

↓

5 — EXECUTE

A separately authorized tool executes the approved query against the customer's business data under existing enterprise controls.

At query time, Atlas itself requires no connection to SAP.

Not a concept. A working system.

From the current NEXEME Atlas reference tenant:

295,000+ tables mapped

13.2 million fields mapped

1.7 million+ foreign-key relationships

1.9 million+ join paths precomputed

~506,000 check-table values indexed

~90 minutes complete artifact rebuild

<200 ms typical Atlas context query

Zero SAP load from Atlas at agent query time

These are measurements from the current reference build rather than estimates.

A generic AI model may know that certain SAP tables commonly contain customer or billing information.

But that doesn't prove:

those are the right objects in your landscape;

your organization uses them in the expected way;

a custom object hasn't changed the picture;

the proposed fields actually exist;

or

the proposed joins are valid in your system.

NEXEME Atlas starts from a different source:

Your SAP system itself.

The Knowledge Artifact is deterministic, inspectable, versioned and rebuildable.

When your landscape changes, rebuild it.

Your agents can reason from a representation of the SAP landscape that actually exists.

FROM STRUCTURE TO BUSINESS MEANING

Finding the table is only half the problem.

Consider our original question:

“List my top 3 customers by revenue with their city and country.”

Atlas may be able to establish exactly which objects, fields, relationships and coded values exist.

But business questions contain another kind of knowledge.

What does this company mean by revenue?

Which customer role should be used?

Which transactions should be excluded?

Those answers can depend on the way an individual organization uses SAP.

This is where NEXEME Atlas is heading next: combining verified SAP structure with a customer-specific semantic layer that captures how the organization understands and uses that structure.

The objective is:

Business language

↓ 

Customer-specific meaning

↓ 

Verified SAP structure

↓ 

Grounded AI reasoning

This is the same translation problem that has existed between business users and SAP technical teams for decades.

AI creates an opportunity to make part of that knowledge machine-readable.

Your SAP context stays inside your environment.

Atlas was designed for enterprises where SAP is a system of record.

Read-only extraction

The standard Atlas build requires display-level access. No write authorizations are required.

No transactional or master business records extracted into Atlas

Atlas maps how the SAP landscape is structured. Configuration/reference values may be included because they help explain that structure.

Potentially personal DDIC author identifiers can be pseudonymized or removed.

Runs inside your infrastructure

The build process, Knowledge Artifact and query server can all operate within your network boundary.

No SaaS dependency

No cloud service is required for Atlas to build or serve the Knowledge Artifact.

No telemetry or phone-home dependency

Atlas can operate without sending usage information to NEXEME.

No SAP connection at agent query time

Once built, AI applications query the Knowledge Artifact rather than connecting to SAP for Atlas context.

Air-gap capable

Atlas can operate without external network connectivity.

MODEL AND AGENT INDEPENDENT

NEXEME Atlas doesn't need to choose your AI for you.

The customer chooses the AI model, agent framework and authorized data-access tools appropriate to its environment.

Atlas provides the SAP context layer those systems can consume.

The current interface uses purpose-built tools exposed through the open Model Context Protocol (MCP), allowing agents to retrieve Atlas knowledge without arbitrary access to the underlying Knowledge Artifact.

More than 30 years between SAP and the business.

NEXEME was founded in Singapore by an SAP practitioner whose experience spans:

SAP R/2 → R/3 → ECC → S/4HANA

Much of that career has involved bridging two worlds.

Business users understand what they are trying to achieve.

SAP specialists understand how the system actually works.

The difficult part is often translating between them.

Enterprise AI is creating a new version of the same challenge.

NEXEME Atlas grew from one question:

If AI agents are going to work with enterprise SAP systems, how will they learn what each organization's SAP system actually means?

Atlas is being built to answer that question.

Put NEXEME Atlas against your own SAP questions.

NEXEME is currently looking for a small number of organizations exploring AI agents, GenAI or intelligent automation around SAP.

I’m particularly interested in environments where:

SAP has been configured and customized over many years;

AI teams need experienced SAP people to explain how the company's implementation actually works;

or AI applications need to turn business questions into reliable queries against SAP-derived business data.

A useful starting point is simple:

Give us real business questions your users want AI to answer.

Then test how an AI agent performs without customer-specific Atlas context — and with it.

[Contact the founder: founder@nexeme.ai]

No sales team. You'll speak directly with the person building Atlas.