Concept card

SAP Datasphere as the grounding source for Joule and agents

As of 2026-09-25

Edited by Cédric-Olivier Mâry, sole editor and publication director — 27 years in the SAP ecosystem. How these pages are produced, and where automation is used.

What is SAP Datasphere as the grounding source for Joule and agents?

Datasphere grounds Joule three different ways — navigational answers inside Datasphere itself (GA since Q1 2026), Analytical Insights over Catalog-exposed analytic models, and Joule Studio agents anchored in the SAP Knowledge Graph and SAP Domain Models — and none of them is document grounding on the HANA vector store. Confusing the three explains most "Joule doesn't know my data" tickets.

Three groundings, not one

"Ground Joule on Datasphere" is used loosely to mean at least three different mechanisms, and the fix for a bad answer depends on knowing which one is in play.

1. Joule inside Datasphere (navigation and task execution). SAP's Q1 2026 Business AI release highlights state that Joule is now generally available in SAP Datasphere, letting users navigate the platform, get answers about spaces and objects, and execute administrative or space-manager tasks in conversational language. This Joule reads Datasphere's own metadata and UI state — it is an in-product assistant, not a business-question answering engine, and it is base functionality, not a premium add-on.

2. Analytical Insights over exposed models (Just Ask and Joule). A second, business-facing path runs through the analytic models a Datasphere space owner exposes to Just Ask and, through it, to Joule's Analytical Insights. This is where the semantic layer — hierarchies, calculated measures, currency conversions, business-friendly labels — becomes the vocabulary Joule reasons with. A model with an unclear hierarchy, an ambiguous label or a missing description degrades the answer even when the underlying data is correct; SAP's Q3 2026 SAC release adds a model readiness check precisely to catch this class of problem before end users do.

Why it matters

Treating "Joule doesn't know my data" as one problem wastes time: the fix is a role check, a catalog edit, or a Knowledge Graph exposure decision depending on which of the three grounding paths is actually broken.
Joule-in-Datasphere navigation is free, base functionality (GA Q1 2026); catalog AI assistance and Joule Studio agent grounding sit on separate, premium or preview tracks — a scoping conversation that skips this distinction under- or over-prices the engagement.
SAP Domain Models move from Early Adopter Care to planned GA in Q3 2026: an agent architecture decided today should assume that grounding channel gets more capable, not stay static.

Key points

Path 1 — Joule inside Datasphere: navigation, answers and task execution on Datasphere's own metadata; GA since Q1 2026; base functionality.
Path 2 — Analytical Insights: Just Ask-exposed analytic models feed Joule's business answers; freshness depends on replication flows, coverage depends on Catalog hygiene.
Path 3 — Joule Studio agents: grounded in SAP Knowledge Graph and SAP Business Data Cloud, increasingly in SAP Domain Models (Early Adopter Care now, GA planned Q3 2026).
Document grounding (PDFs, SharePoint, the HANA vector store) is a separate, fourth channel not covered by this card — see C116.
Since July 2026, personal-data columns in BDC data products are auto-tagged in the Catalog, narrowing but not removing the steward's review.
Q3 2026 SAC adds a model readiness check that flags hierarchy, label and semantic gaps before they degrade Joule's answers.
A private, ungoverned Datasphere view is invisible to Joule by design — exposure is a deliberate governance decision, not a default.

Terms used on this page

Joule Studio: SAP's managed environment (announced May 2026) for building and running the full lifecycle of AI agents, grounded in the Knowledge Graph, BDC and Domain Models.
SAP Knowledge Graph: SAP's semantic backbone connecting business entities and their relationships, used to ground Joule agents in structured business context.
SAP Domain Models: Domain-specific models trained on SAP code, customer data, metadata and business processes; Early Adopter Care today, GA planned Q3 2026.
Analytical Insights: Joule's business-question answering path over analytic models exposed to Just Ask in SAC/Datasphere.
Replication flow: The Datasphere mechanism that brings non-SAP-managed data in via initial load, delta/CDC, on a scheduled or continuous cadence.
Model readiness check: SAC Q3 2026 feature that evaluates an analytic model for hierarchy, label and semantic issues before it is used by Just Ask or Joule.

Sources

Full card available to members. What the full card adds: the full decision framework · the common pitfalls and their fix · the cheat sheet · the facts worth quoting.

Open in the app →