Prepare and maintain a connected SAP system
Before a data source can answer anything, the AI Context Hub has to catalog what the connected SAP system can do. Ingestion crawls the system and records its services, operations, and fields, with a description of each, into a searchable catalog. This is an occasional, once-per-system step, configured on the system rather than on a data source, and it is usually done by an administrator.
SAP-side prerequisites
Most source kinds ingest with no extra setup, but some need configuration on the SAP system first. Meeting these before you ingest avoids a source kind failing to catalog.
This assumes the system-wide requirements are already in place, in particular a Neptune DXP - SAP Edition instance and the imported AI Context Hub API transport. See System requirements.
- API Factory sources (function modules)
-
The ingestion user needs a suitable API Factory policy, or the source’s Visible to all switch enabled, and the dynamic service node must be active. Without the right policy, the orchestration engine cannot list what the system offers, and ingestion of that kind fails.
- Neptune DXP - SAP Edition APIs and SAP Integration Hub APIs
-
Each API must publish an OpenAPI specification. An API without one is not ingested. The SAP Integration Hub runtime must be installed for its APIs to be discovered.
For what each source kind is, see Sources you can connect.
Procedure
Ingest the system
You configure ingestion on the connected remote system, then run it.
-
In the Remote Systems tool, open the remote system, and on its AI Context Hub tab, enable SAP Engine.
-
Choose what to ingest:
-
Select the application components to ingest in the Application Components Browser. Each source kind toggles on or off independently, which controls what the orchestration engine, and so an agent, can reach.
-
Add sources that are not in the browser with Manual Additions (Advanced): an OData service through Browse OData Services, or a Neptune DXP - SAP Edition API by its API path.
Ingesting a whole SAP system can catalog far more than an agent needs and take a long time. Select only the components your data sources will use. Selecting a component pins it, and a component pin resolves dynamically at ingestion, so it also picks up objects added in SAP later. Narrowing a business object to specific methods ingests those methods on their own and drops the object-level pin, so that object no longer widens automatically. -
-
Set the Embedding Model, used to embed the ingested services so a request can be matched to them by semantic search. This model and the enhancement model both come from the Model tool in Naia Agent Studio.
-
Turn on Enhance with AI to have the orchestration engine write richer service descriptions, and select the model that writes them. Better descriptions improve how well a request matches the right operation.
-
Set how long an already-ingested part of the catalog is reused before it is fetched again.
-
Ingest the configured services.
Result: The orchestration engine builds the catalog. The system’s services and operations are then available to grant on a data source’s SAP Services tab. See Grant SAP services and operations.
Ingestion is manual: it does not run on a schedule. You start it, re-run it when the system changes, and you can delete the ingested metadata or preview the ingested SAP services from the same place. Between runs, data sources answer from the existing catalog without touching the system’s metadata.
Reuse a data source on another system
You build and test a data source on a development system, then move it to a production system. Because a data source depends on the catalog of the SAP system behind it, moving the data source is only part of the job: the target system needs its own connection and its own catalog.
-
On the target system, set up the remote system that connects to its SAP system.
-
Deploy the data source to the target system.
-
Ingest the target’s SAP system, or re-ingest it if it has changed.
-
Open the deployed data source and confirm its granted services and operations resolve against the target catalog.
If the target SAP system exposes different or missing metadata, some granted operations may not resolve. Review the data source after deployment and correct any that do not. -
Test a request query in the Playground on the target.