Can Now Assist in Virtual Agent generate answers from non-KB tables (e.g., Incident)?

[marloii]

Is it possible for Now Assist in Virtual Agent to use data from non-Knowledge Base tables (such as the Incident table) to generate responses?

I was previously able to do this during testing in the Xanadu release. However, when trying again in a Yokohama lab instance, it no longer works.
In Assistant Designer, the incident search source shows the following tooltip:

"This search source will appear in search results, but the LLM won't use it to generate responses. LLMs can only use Knowledge Bases, External Content Connectors, or Catalogs."

Has this behavior changed in recent releases?

If so, is there any supported workaround to allow LLM-generated answers from other tables?

[Yep]

Thanks for the reply Brian! Hope this answer your questions well.

We have a custom table where we store our 3rd party application role info (e.g. Active Directory groups. Really any app the company has we store rights within and installs). Users have a hard time knowing which role to pick when filling out access requests so were trying to recommend them using Now Assist inside the Virtual Agent on Employee Center.

In our table we have a few thousand roles and 3 key fields: Group/Role Name, Description, Business Application (reference to APM). I'm having a hard time thinking of access examples but mostly along the lines of "I need X thing in X app".

I have tried AI Search but results are mixed when doing a RAG call in a skill so looking to explore other options.

[rpriyadarshy]

 @Yep here is just a thought. Create an  "Access Role Recommendation” AI Agent" 

Invoke This Agent when User says- What Role is needed for developer access in SAP Application. 

Application- SAP 

Access Type - Developer

 The Tool in Agent will fetch the details using script or Sub-flow from Your Custom Table. It May pull >1 Records. Now Send this output array to LLM for precise result .

Regards

RP

[Brian Bakker]

@Yep - Have you enabled Semantic vector search in AI Search? Use the Semantic Index Configuration form to setup semantic indexing on your 3 key fields Group/Role Name, Description, Business Application. Once indexed with the semantic embeddings it will enable DPR (Dense Passage Retrieval) for improved recall compared to just keyword search. Then you enable Hybrid search in AI Search and you Manage hybrid search in search applications, not the search profile.

You also have these options when you Add a search retrieval to an AI agent or when you Add a retriever in your Now Assist Skill where you can enable Hybrid search.

Finally, please be aware of the system property [glide.ais.query.hybrid_allowed_tables], as you will need to add additional tables, such as your custom table to this system property for Hybrid Search, and a user with the "ais_admin" role is able to update this property.

Let me know if this configuration has improved your search results on your custom table.

Regards,

Brian