Jonatan SkopekData and AI Consulting
Services

AI Agents and Machine Learning on SAP Data Platforms

Technologies

SAP HANA CloudSAP DatasphereSAP BDCJoule StudioMCPHANA ML (PAL / APL)Databricks

AI agents can answer questions directly on the data in SAP HANA, Datasphere, BW/4HANA and the Business Data Cloud: they understand the data model, formulate queries, have KPIs, forecasts and ML models calculated in the database and put the results into business context. I support you from selecting suitable analytics use cases through the architecture decision to implementation – including the MCP server, the authorisation and audit concept and a semantic description of your data model so that the agent delivers reliable results.

Architectures

Four ways to connect agents to your SAP data

Which architecture fits depends on cost, performance, data privacy, scalability and the required quality of results. I assess these criteria together with you and implement the right option.

Cloud agent with MCP server and SQL access to SAP HANA

A cloud agent such as Claude Code or Microsoft Copilot works directly on HANA, BW/4HANA or a Datasphere space via a custom MCP server. This offers maximum flexibility: any SQL, calculation views and machine learning with PAL and APL – calculations are pushed down into the database. I design authorisations, audit and protection against unwanted queries from the start.

User & AgentBusinessUserCloud Agent(Claude / Copilot)CloudLLMTools (MCP)CustomMCP ServerSAP BTPor VMAuthorization& AuditInterfaceJDBC /hdbcliSQL &Calc ViewsPAL / APL(ML)DataHANA Tables& ViewsBW/4HANADatasphereSpaceSAP HANA Cloud · SAP HANA

Cloud agent with MCP server and OData access to Datasphere

The agent uses the Datasphere consumption APIs and thus works on the modelled business layer: analytic models, business names and KPI definitions. If the agent accesses data on behalf of the user, data access controls apply as well. The MCP server can run on SAP BTP, a VM or in SAP Cloud Integration.

User & AgentBusinessUserCloud Agent(Claude / Copilot)CloudLLMTools (MCP)CustomMCP ServerBTP, VM orCloud IntegrationUserPropagationInterfaceODataConsumption APIAnalyticModelsData AccessControlsSemantic LayerSpacesViews &DimensionsBusinessNamesSAP Datasphere · SAP HANA Cloud

Joule agents on Business Data Cloud data products

The SAP-native way: agents built in Joule Studio use curated data products with built-in semantics and governance, while SAP takes care of operations and scaling. I assess with you which capabilities are available today, which are announced and how this path fits into your roadmap.

User & AgentBusinessUserJoule /Joule WorkA2A / MCP(announced)AgentsJoule StudioAgentsJoule StudioRuntimeSAP AIAgent HubData ProductsSAP-managedData ProductsCustomData ProductsSemantics &GovernanceSAP Business Data Cloud

Local model with the hana-ai MCP server

For sensitive data, prototypes and architecture tests: a locally hosted language model uses the MCP server from the HANA AI Toolkit following the SAP reference architecture. Query results are not sent to any AI provider, every tool call can be traced in the HANA audit and there are no token costs.

Local Machine · User & AgentBusinessUserBionic(Agent Harness)Local LLM(LM Studio)Local Machine · Tools (MCP)hana-aiMCP Serverlocalhost+ TokenAudit viaSession ContextInterface & Datahdbcli /SQLHANA ML(PAL / APL)Semantic Catalog(Vector)SAP HANA Cloud

My principle from projects and hands-on tests: the quality of a data agent depends less on the model than on the division of labour. The model plans and interprets, MCP tools translate and safeguard, the database does the calculations – and a well-maintained semantic description of the data model often helps more than a bigger model.

From the blog

Insights from practice

An in-depth article on this topic – with an architecture comparison and a hands-on test.

Answer of the local AI agent in Bionic: tables with RMSPE per store and the sales forecast.
Agentic AI · SAP Data & Analytics

Agents on SAP data: four approaches and when to use which

AI agents are expected to answer questions directly on data in SAP HANA, Datasphere, BW/4HANA and the Business Data Cloud. There are several architectures for this, each with a different price in cost, data privacy and quality of results.

SAP HANA CloudDatasphereBW/4HANABusiness Data CloudJouleMCP
Article in GermanRead article
References

Relevant projects

Selected projects from my practice in this service area.

Automotive & Manufacturing

Transformation of Existing BW Systems into a Central Hybrid BW/4 and Datasphere Platform

Lead Architect and Technology Expert

  • Development of guidelines and procedural models for the transformation
  • Development of platform concepts for BW/4HANA (development guidelines, data architecture) and SAP Datasphere (space concept, development guidelines, authorization concept)
  • Development of the target architecture at platform and application level
SAP BWonHANASAP BW/4HANASAP HANASAP Datasphere+2
Automotive & Manufacturing

Discovery and Target Architecture for the Transformation of BW on HANA Systems to Datasphere

Architect

  • Analysis of the as-is architecture of a BW on HANA system at platform and application level, incl. AI-supported creation of detailed application profiles
  • Development of various options for the target architecture and evaluation via capability matrix
  • Validation of the architecture options via PoT implementations
SAP BWonHANASAP DatasphereSAP Analytics CloudSAP BDC

Do you have similar requirements?

Talk to me about your project. I will analyse your situation and show you possible ways forward.

Get in touch