Rovo agent knowledge base from PDFs: Google Drive/OneDrive vs Confluence approach

[Petr Sýkora]

Hi,

I want to build a Rovo agent in Rovo Studio that answers based on a specific set of documents. This works well when the documents are in Confluence, because I can select a specific space or a parent page as the knowledge base.

Now we have another use case: a knowledge base made up of multiple PDFs of various sizes. I wanted to use the Google Drive and OneDrive connectors for this. However, in this case I can’t select a dedicated folder as the source for the Rovo agent’s knowledge base. Is there a way to do that?

As an alternative, I’m considering creating an app that extracts text from the PDFs and stores it in Confluence as pages. That would also (hopefully) work for PDFs larger than 10 MB (which seems to be the connector limit).

What would you recommend? Has anyone worked on a similar use case?

[Ryan Boyd · Accepted solution]

I tackled this by embedding the PDF in a confluence page and using that page as the single source of knowledge for the agent. It cut out the need to do something complex with pulling the information from the PDF into confluence and worked rather well. 

[Giuseppe Torraco]

Embedding PDFs in Confluence can work as a short‑term workaround for a small number of documents.
However, it doesn’t scale when dealing with hundreds or thousands of PDFs, which is common in enterprise contexts.

At scale, this approach creates issues around:

• Maintainability (one page per PDF, updates, versioning)
• Retrieval quality (large, unstructured content reduces agent accuracy)
• Governance (no real KB structure, taxonomy, or quality standards)

In our experience, PDFs work better as source material, while the actual KB for Rovo should be structured, AI‑native Confluence content designed for retrieval, governance, and long‑term evolution of agents.

Fine for quick wins—insufficient for production-grade, enterprise Rovo agents.