Cursor's codebase semantic search relies on vector embedding technology under the hood, converting code into high-dimensional vectors stored in a local vector database, and retrieving the most semantically similar code snippets via cosine similarity.