How does Rovo deal with large amounts of retrieved content?

Hi @AGASTYA SHARMA 

There isn't any official documentation that explains the internal Retrieval-Augmented Generation (RAG) pipeline in detail.
For example, Atlassian hasn't shared whether Rovo summarizes retrieved content before sending it to the LLM, how many documents or chunks are included in the prompt, or how it handles the model's context limits.

Rovo doesn't use your entire knowledge base to generate an answer. Instead, it searches across Jira, Confluence, and connected third-party applications, retrieves the most relevant content that the user has permission to access, and generates an answer based on those results. It also includes citations so you can trace the answer back to its original sources.
                                                             The Atlassian Engineering team has also explained that Rovo ranks content using several relevance signals, including metadata, recency, popularity, authority, contributor information, and user affinity. These signals help identify the most relevant content before an answer is generated, rather than attempting to use every matching document.

So, based on the publicly available information, it's reasonable to say that Rovo retrieves and ranks the most relevant content before generating a response.

Official references:

• Atlassian Engineering – Unraveling Rovo Search:
  https://www.atlassian.com/blog/atlassian-engineering/unraveling-rovo-search

• Rovo Search documentation:
  https://support.atlassian.com/rovo/docs/search/

• What is Rovo?:
  https://support.atlassian.com/rovo/docs/what-is-rovo/