Repository navigation
Adapt to use VECTOR_EMBEDDING Function (Vector) and other new AI Core features within the HANA Database #2810
Description
There are now Database features to do AI Core calls such as creating ebeddings directly within the database. This can be done more efficiently that at the Node.js layer with the Cloud SDK. We should look at where we can utilize this new functionality.
Also this help link for setting up the connection to AI Core https://help.sap.com/docs/hana-cloud-database/sap-hana-cloud-sap-hana-database-vector-engine-guide/creating-text-embeddings-with-sap-ai-core
Activity
docs(#2810): avoid PEM markers in PSE example to satisfy gitleaks
docs(#2810): HANA-native AI Core embedding setup + grant CREATE PSE
Merge pull request #2844 from sap-tutorials/feat/2810-hana-aicore-emb…
Merge pull request #2846 from sap-tutorials/feat/2810-hana-aicore-pse…
Merge pull request #2847 from sap-tutorials/fix/2810-pse-trust-admin
Merge pull request #2848 from sap-tutorials/fix/2810-pse-admin-not-co…
Merge pull request #2849 from sap-tutorials/docs/2810-external-privil…
jung-thomas commentedon Oct 19, 2026
Status: basic setup DONE — HANA↔AI Core connection live and verified on DEV + PROD
The first phase (connect the existing AI Core instance to HANA and prove
VECTOR_EMBEDDINGfrom the DB layer, before any rewrite) is complete.Proven end-to-end (as the app's own container
_RTidentity)SAP_GXY.20250407)TUTORIALS_AICOREremote sourceGET_REMOTE_SOURCE_AI_MODELSFinal architecture
CREATE PSE TUTORIALS_AICORE_PSE+CREATE REMOTE SOURCE TUTORIALS_AICORE ADAPTER "sapgenaihub"using the dedicatedhana-ai-corex509 service key (deploymentda6d9c5e3fb50c3d). PSE + remote source are shared DB objects, not container artifacts.GRANT EXECUTE ON REMOTE SOURCE TUTORIALS_AICORE TO "<schema>::external_privileges_role"— the broker's external-object role; rebind-proof, inherited by all binding users. Not an_RT-user grant, notdefault_access_role, no grantor service.Shipped
Runbook
docs/developers/operations/hana-aicore-embeddings.md(both-env SQL, tiers, PROD marked done, auto-embedding + wider-AI-surface appendices) + design doc +CLAUDE.mdlink. PRs: #2844, #2846, #2847, #2848, #2849 (all merged to DEV). The runbook records the failed approaches (.hdbgrants,CREATE PSEas a privilege name,default_access_role+TRUST ADMIN) so they aren't re-tried.Follow-up (next phase — the actual efficiency payoff)
Move embeddings currently generated at the Node/CAP layer into the DB using HANA's auto-embedding mechanisms (
CREATE_VECTOR_EMBEDDINGSprocedure, scheduler job, orBEFORE INSERT/UPDATEtrigger — NCLOB-aware;GENERATED ALWAYS ASis NVARCHAR-only and doesn't fit the gzip/NCLOB content model). Candidate first target + wider AI surface (AI_TEXT_COMPLETION,CROSS_ENCODE,AI_TABULAR_PREDICTION) noted in the runbook appendices.Ops note
Both
hana-ai-coreservice keys (DEV + PROD) should be rotated — their x509 private keys were handled during setup.Merge pull request #2851 from sap-tutorials/fix/2810-docs-sidebar-han…