Can Virtual Agent access documents attached to sys_user record using Agentic flows?

[Mohamed_009]

Hi Community,

I’m exploring the capabilities of Agentic flows with Document and Visual Insights (DVI Agent) and would like to confirm a use case.

Use case:
We have documents attached to the User Profile (sys_user record), for example:

• ID proof

• Offer letter / employment documents

• Address proof

• Other employee-related documents

My question is:

If a user interacts with Virtual Agent and enters a prompt like:

• “What is my employee ID?”

• “What is my joining date from my offer letter?”

• “Show details from my attached documents.”

Can the AI Agent automatically:

1. Fetch the attachments from the logged-in user’s profile (sys_user record)

2. Understand those documents

3. Answer the user’s question using the information from the attached documents?

In short, can Virtual Agent + Agentic flows use documents attached to the user record as context and answer questions dynamically?

If yes, could you please share:

• Recommended architecture / approach

• Whether this should be implemented using Agent Studio / AI Agent orchestration

• Any best practices or limitations

Thanks in advance for your guidance!

[Dr Atul G- LNG]

Hi @Mohamed_009 

https://www.servicenow.com/community/now-assist-articles/building-agentic-flows-that-can-understand-...

CC:

@Loic1 

[Dr Atul G- LNG]

Ho @Mohamed_009 

Did you get chance to above link,

[rpriyadarshy]

@Mohamed_009  few points i was thinking.

End User will not have any kind of License/role i assume. So, as end User i don't think IN any way Service-now Security/ACL will Allow to access those data with End User Permission Roles. As Now Assist/Agents will not bypass record security or attachment permissions.

AI Agents cannot automatically “understand” offer letters or ID proofs unless those documents are explicitly ingested into a RAG-capable source (e.g., Knowledge Base or AI Search index) with Proper Permissions/Roles.

you must build a custom Agentic flow: fetch attachments via Flow/API, extract text , and pass it as controlled context to the LLM.

Trigger that Agentic Flow from VA Topic.

Regards

RP

[Mohamed_009]

@rpriyadarshy 

Thanks for the explanation, this really helps.

Could you please let me know if there is any solution document, official documentation, blog, or reference links that explain this approach (Agentic Flow + attachment extraction + Virtual Agent)?

I would like to review any recommended architecture or implementation guidance before starting the design.

Any links or references would be greatly appreciated. 🙏