Data Source Configuration

Data Source Management is an application that enables SAP data sources to be used in AI Agent conversations. It supports the configuration of system destinations, discovery of available data sources, management of their replication for Retrieval-Augmented Generation (RAG), and assignment of selected sources to AI Agents.

When properly configured, structured business data from SAP systems can be securely discovered, optionally replicated, and made available to AI Agents at runtime.

Overview of the Process

The overall process consists of the following steps:

1. Register SAP data sources
2. Create a destination with data source discovery enabled
3. Discover available data sources
4. Manage discovered data sources and replication settings
5. Assign data sources to AI Agents

Data Source Scenarios Management

Before data sources can be discovered, they must be registered in the SAP system.

To enable data source discovery:

• Data sources must be registered in table /SKYBFR/YAICDSNM
• Only registered data sources can be discovered by the system

Destination Management

Step 1: Open Destination Management

Navigate to the AI Agent group and open the Destination Management application.

Step 2: Create a New Destination

Click the Add (+) button to create a new destination.

Provide the required connection and authentication details.

Enable Data Source Discovery for the destination.

Save the destination.

Once the destination is created:

• The system automatically runs data source discovery.
• The destination overview displays: last Run timestamp and number of discovered data sources
• A manual refresh option is available to trigger discovery again.

From the destination overview, you can navigate directly to the Data Source Management application using the redirection button.

Step 3. Registering Data Sources for Discovery

This section describes how to register existing SAP data sources so that they can be discovered and later used by AI agents via a configured destination.
The registration process does not create new data sources; it only makes existing objects available for discovery.

Prerequisites

1. The data source to be registered already exists and is active in the SAP system
2. The user has authorization for: Transaction SM30 and maintenance access to table /SKYBFR/YAICDSNM
3. A destination is configured for Data Source discovery.

Procedure

Execute transaction SM30.

Enter table name /SKYBFR/YAICDSNM and choose Maintain.

Choose New Entries.

Enter the required information:

Object Name: Technical name of the existing object (for example, a CDS view or service name).
Description: Short functional description of the data source.

Save the entry.

Result

• If the specified data source exists in the system, it is discovered on the configured destination.
• The data source becomes available for the later use by AI agents.
• If the data source does not exist, it is not discovered.

Data Source Management Application

Step 4. Open Data Source Management Application

To manage discovered data sources, navigate to the AI Agent group and open the Data Source Management application.

Main Page Overview

The main page displays all discovered data sources across destinations and includes the following elements:

1. Destination Name
Displays the destination along with the number of data sources discovered for it.

2. Run Data Source Discovery
Manually triggers data source discovery for the selected destination.

3. Switch Destination
Allows switching between different configured destinations.

4. Search, Filter, and Sort Area
Enables searching, filtering, and sorting of data sources.

5. Adding datasources to AI Agent

6. Data Source List
Each data source entry shows:

• Name
• Type (Table, CDS, View, Other)
• Description
• Replication status (On / Off)
• Option to directly assign the data source to an AI Agent

Step 5. Data Source Details View

Selecting a data source opens the Data Source Details view, which provides detailed technical and replication information.

1. List of Data Sources

2. General Information

Type: (Table, CDS, View, Other)

Description: data source description for SAP system

Replication status (enabled or disabled)

Add to AI Agent: option to add data source to the selected AI Agent

Data Scope

The Data Scope section limits how many records of a data source are made available to AI Agents. It is shown for every data source, directly above Replication Configuration.

Without a limit, every full replication run reads the whole data source. A limit is useful for large tables, where a full run takes long and vectorises far more data than an agent needs.

Max Records is the largest number of records that replication loads for this data source. Leave the field empty for no limit; the section then shows Unlimited.

Records are selected in key order, so the same limit selects the same records on every run.

Note

The limit applies to full replication runs. With Delta replication it applies to the initial full load; later runs are not limited and add only the records changed since then, based on the selected Date Fields.

Setting a record limit

Step 1. Open the data source and choose Edit in the Data Scope section.

Step 2. Enter the number of records in Max Records, or clear the field to remove the limit, then choose Save. The field takes a whole number of up to nine digits.

An entry that is not a whole number greater than 0 shows an error on the field, and Save is not available until you correct it.

Caution

If the data source exposes no key fields, the dialog warns that a limit may select different records on each run. You can still save the value. For a stable selection, expose a key field in the view, or have the key fields maintained for the data source in the backend configuration.

When a run reaches the limit

If a replication run stops at the limit, a Record limit reached tag appears under Last Replication Run. Choose the tag for the explanation and what to do about it.

The records beyond the limit are not replicated, so they are not available to AI Agents, and Delta runs do not add them later. To make more records available, increase Max Records. With Delta replication, run Reinitialize Delta afterwards, so that the higher limit is applied in a new full load.

In the Replication Log, a run that stopped at the limit shows an orange icon in front of the number in Rows Replicated. The run itself still counts as a success: the records it replicated are complete and in use.

3. Replication Configuration

The replication section allows configuring and monitoring data replication for the selected data source.

Replication is required for data sources that should be stored and used by Retrieval-Augmented Generation (RAG).

Enabling Replication

To enable replication:

Switch the Replication toggle to On and select the Replication Type:

• Full Replication– the complete data set is replicated
• Delta Replication– only changed data is replicated (if supported by the data source)

Note

When replication is initialized, a dedicated space for vectorized documents is automatically created. Within this space, you can monitor the vectorization status and review the documents that are available for use by AI Agents.

Full Replication

Full replication synchronizes the entire data set from the source system during each replication run. No change detection is applied.

Type: Specifies the replication mode for the data source

Frequency: Defines how often replication is executed.

Last Replication Run: Displays the timestamp of the most recent replication execution.

Run Replication button: triggers replication manually for the selected data source.

Edit button: Allows changing replication settings such as type and frequency.

View Replication Logs: Opens a list of replication runs with status and execution details.

Note

During full replication, newly fetched data is vectorized as an independent document while the currently active document remains available to AI Agents. Only after the new document has been fully vectorized is it published and replaces the previous version in RAG.

This approach ensures continuous data availability and prevents gaps during the replication process.

Delta Replication

Delta replication enables incremental data transfer by detecting changes since the last replication run.

Prerequisites

• Delta replication is only available for data sources that expose a suitable date field.
• The date field must represent the last change timestamp of the record.

Without a date field, incremental change detection is not possible.

Preparing Data Sources for Delta Replication

If delta replication is required, ensure that the data source exposes an appropriate date field:

• Tables: Create a new Data Source that includes or joins a relevant date column.
• Views: Extend or adjust the view to expose a date field suitable for change tracking.
• CDS Views: Add or expose a date field or parameter that reflects the last modification timestamp.

Note

The selected date field must indicate the last update or change of the data record.
Fields representing creation date only are not suitable for delta replication.

Delta Field Selection

When configuring delta replication, select one or more Date Fields that will be used for delta detection.
Ensure that the selected fields accurately reflect changes made to the underlying data.

Reinitialize Delta

After delta replication is configured, the Reinitialize Delta option becomes available.

Reinitializing delta:

• Resets delta tracking for the data source
• Causes all data to be fully replicated again during the next replication run

This option is typically used after:

• Structural changes to the data source
• Corrections to delta field configuration
• Data inconsistencies

View Replication Logs: Opens a list of replication runs with status and execution details.

Filters

Filters narrow a data source to the records that matter. Only records that match the filters are replicated and made available to AI Agents, so an agent can work with the orders of one sales organization instead of the whole table.

Filters belong to the replication configuration: you set them when you switch replication on, and change them with Edit in the Replication Configuration section. Filters decide what replication loads, so a data source that is read directly, without replication, is not filtered.

Note

Filters are applied before the record limit. Records are first narrowed by the filters, then capped by Max Records of the Data Scope section.

Setting filters

Step 1. In the Replication Configuration section choose Edit, or switch Replication on to configure it for the first time.

Step 2. In the Filters block choose Add Filter, and pick the fields to filter in the Select Fields dialog. Only fields the data source exposes are offered.

Step 3. For each field, open its Conditions and define them. A condition is a guided row with an operator and a value in the format of the field; there is no free-text SQL. A condition can also exclude records, or test a field for <empty>.

Step 4. Choose Save.

A filter belongs to one field. When a filter has several conditions, a record has to match at least one of them; when several fields have filters, a record has to match all of them.

Clear Filters removes every filter at once; the delete icon of a row removes the filter of that field. With no filters, the section shows None. All records are in scope.

When the filters take effect

A changed filter applies from the next replication run. Records already replicated are not filtered again until that run.

Caution

With Delta replication a run reads only the records changed since the last run, so data already loaded keeps the selection made under the previous filters. Use Reinitialize Delta to apply the new filters to all records.

The section shows the filters that are saved, under Applied Filters. If replication has not run since you changed them, they are not yet the filters the replicated data was selected with.

Messages you may see

• The filters are too long to be saved. Remove some conditions. The filters do not fit the space stored with the data source. Use ranges or patterns instead of long lists of single values.
• Invalid Condition Value. The value does not match the format of the field. Enter it as the field stores it, for example a date in the field’s own format.
• Custom condition. A condition saved earlier that cannot be shown as filter rows. It stays in use until you add a filter, which replaces it.

Caution

If a field used in a filter is later removed from the data source in the SAP system, replication still starts, but the vectorized data cannot be built. Remove the filter on that field, or expose the field again.

Filters decide which records an AI Agent can see; they are not an authorization check. To restrict records by the organizational units a user is authorized for, see Organizational Authorization for Data Sources.

4. Fields Information

The Fields section displays detailed metadata about the structure of the data source.
For each discovered field, the following information is shown in a table:

• Field
• Data Element
• Domain
• ABAP Type
• Length / Decimals
• Short Text
• Medium Text
• Long Text
• Key
• Lowercase
• Sign

This information helps assess the structure and suitability of the data source for AI Agent usage.

5. Parameters (CDS Data Sources)

If the selected data source is a parameterized CDS view, a Parameters section is displayed in the details view.

The parameters table shows the following information:

• Parameter
• Data Element
• Type
• ABAP Type
• Length / Decimals

If a data source does not define parameters, this section is not displayed.

Step 5. Searching, Filtering and Sorting

You can search, sort and filter this data using the following buttons:

and decide which filds should be visible:

Step 5. Assigning Data Sources to AI Agents

From both the Data Source Management list and the Data Source Details view, data sources can be directly assigned to AI Agents.

On the main application view you can select more than one data source and add them to AI Agent.

You can also enter data source details view and add it directly from there:

When a data source is assigned to an AI Agent, a corresponding tool is automatically added to that agent. The tool is named after the data source and includes information about its replication status.

All such tools are created with the type “datasource”, enabling the AI Agent to access and use the replicated data during runtime.

Step 6. Managing Input and Output Structures on the Tool Level

This section describes how Data Sources and Business Tools are executed by the AI Agent, how data is passed into the tool, how results are returned, and how outputs are stored for later use.

Data Sources and Business Tools follow a common execution model when used by an AI Agent:

• Input data is passed to the tool
• The tool is executed
• Output data is returned in a structured format
• The result is stored in the AI Agent’s memory and can be reused

While creating new tool select type: datatsource or business and you will be able to establish necessary input and output structures.

Input Structure
Describes the data required to run the tool.

• For Business Tools, input values are provided by the AI Agent or previous steps. You must provide the input json structure to enable Business tool to be used by MCP server or A2A interface.
• For Data Sources, input fields are automatically populated based on metadata (for example, table parameters, views, or filters). No manual input mapping is required for metadata-driven data sources.

Output Structure
Defines the format of the data returned by the tool after execution.

• Data Sources return structured data from tables, views, or queries.
• Business Tools return execution results, such as document numbers, statuses, or messages.
  The output structure must match the expected schema to ensure proper downstream usage.

Target Memory Parameter
Specifies where the tool output is stored in the AI Agent’s memory.
Stored results can be:

• Reused by subsequent tools or skills
• Referenced in prompts and logic
• Included in responses to the user

Each tool execution can write to a dedicated memory parameter to avoid data conflicts.

NOTES:

• Data Sources are typically metadata-driven and read-only
• Business Tools require explicit input and output definitions
• Correct memory configuration is required for multi-step AI Agent scenarios

Tool view with the information in which tools input and outpot structures are maintained: