How Does Cursor Use Pinecone for Public Web Document Embeddings?

I’m currently using the Cursor IDE and have encountered some questions I’d like to ask in the forum.

While reading the information on Cursor’s official website, I came across a section that mentions:

“Pinecone sees no code data: Embeddings and metadata of some indexed docs are stored on Pinecone. These docs are fetched from the public web.”
(Source: Security | Cursor - The AI Code Editor)

It appears that Cursor uses Pinecone, a vector database, to store embeddings of certain documents that are fetched from the public web. This seems to be different from the current Web Search feature. I’m curious: which feature or functionality in Cursor actually utilizes this part of the design?