How the AI Context Hub works
The orchestration engine answers a request by working from a catalog of the connected SAP system, not from the system directly. The SAP data itself stays separate from the AI model throughout.
The answer mechanism
Whatever a request asks, and whichever SAP sources it draws on, the orchestration engine handles it the same way, from the catalog it searches to the answer it returns.
The catalog
The catalog is a searchable inventory of everything the connected system can do, built when the system is ingested:
-
Every SAP source is recorded in one uniform model, whether OData, a BAPI, or an API, so the orchestration engine plans and calls them all the same way.
-
Each entry carries a plain-language description, so a natural-language request can be matched to it.
-
Re-ingesting the system rebuilds the catalog whenever it changes.
See Prepare and maintain a connected SAP system for ingestion and Sources you can connect for the source types.
Planning and execution
The orchestration engine works through each request in the same order:
-
It searches the catalog for the operations that fit the request.
-
It plans the steps and fills in their parameters, looking up a code first when the request names an entity by name.
-
It runs the steps against the live SAP system, as the user.
-
It returns the answer as text or a card.
See the Playground to watch these steps for a single request.
The data boundary
The AI model works from metadata only:
-
It reads the catalog, meaning service names, field names, and descriptions, to plan the answer.
-
It never sees the values SAP returns. Those travel straight back to you as the answer.
See Data security and governance to set how strict this boundary is per data source.
Speed and quality
The plan cache
A successful plan is cached and reused for a later request that means the same thing, which is faster. The cache is safe to leave on: when no cached plan is close enough, the orchestration engine plans the request afresh.
See Configure the orchestration engine to control it per data source.
Answer quality
Answer quality depends as much on your setup as on the orchestration engine:
-
The AI model: A more capable model handles complex queries more reliably, and you choose it per data source. See Agents.
-
The instruction: Clear context and presentation rules guide the orchestration engine to the result you intend. See Write the instructions.
-
The catalog: Current, well-scoped ingestion produces better answers than a stale or narrow one. See Prepare and maintain a connected SAP system.
Simple lookups are more predictable than complex, multi-step queries or updates. Test representative queries in the Playground before an agent relies on a data source.