RAG on CDS

Puts the retrieval into one call: a single method that embeds the question through SAP AI CoreSAP AI CoreserviceSAP's AI service on BTP: the runtime for AI assets, and through the Generative AI Hub inside it the access to foundation models, the orchestration service with its templating, masking, filtering and grounding modules, the Prompt Registry and Document Grounding.sap official, btp, searches SAP HANA CloudSAP HANA CloudserviceThe managed database under most of this: CAP's default target, the store Datasphere and Analytics Cloud sit on, and a service with plans and components.sap official, btp's vector engine for the nearest rows and sends them on to the model. The embedding and the similarity search are also exposed on their own, for a handler that wants to keep the steps apart. What gets embedded, and when, is the application's: the README describes no chunking and no embed-on-write.

Reaches a Document Grounding collection through the LangChain vector store interface, so ingestion, chunk storage, embedding and retrieval happen in SAP AI Core and nothing is persisted in the application's database.

Holds chunks, their embeddings and their metadata as CDS entities in the application's own database, and retrieves them by cosine similarity through the LangChain vector store interface.

Answers the generation half only. ragChain takes documents the caller has already retrieved, assembles them into a prompt under a context budget and sends it to the configured provider; embed on the same client produces the vectors, and semanticCache and embeddingDedup reuse earlier answers. The store, the index and the search are not in the package, which is what @saptarishi/cds-plugin-vector-hana@saptarishi/cds-plugin-vector-hanacds-pluginVector storage and similarity search on SAP HANA from CAP, packaged as a plugin so the embedding column and the query are modelled rather than hand-written.independent, cap composes with it to supply.

Answers the storage and retrieval half inside a CAP project: a vector table per @rag-annotated entity kept in sync by the plugin's own CRUD handlers, top-K cosine search with an optional metadata filter, hybrid vector-and-keyword retrieval fused by reciprocal rank, and a RAG helper that assembles the hits into a citation-tagged prompt and calls a chat provider, with a streaming variant that returns the hits before the first token. The embedding and chat models come from a separate provider plugin, and the embedded unit is one row's projected fields.