Codebase indexing

When you open a project, Cursor scans and indexes your source files. Indexing turns chunks of code into embeddings, numeric fingerprints of meaning, so the agent can run semantic search: "where do we check user permissions?" finds the right functions even if they never use the word "permissions". Ordinary text search still exists too; the agent uses both.

Practical details:

• Indexing runs automatically on open and then re-syncs periodically (roughly every five minutes) to pick up changes.
• The status bar shows progress while a scan runs; until the first index finishes, search results are weaker.
• To rebuild from scratch, open the Command Palette and run Reindex. Do this if search seems to miss code that clearly exists, for example after a huge refactor or branch switch.

What gets indexed matters. Files in your .gitignore (like node_modules or dist) are excluded, plus a built-in list of things such as lock files and binaries. Add large generated files, vendored code or data dumps to .cursorignore so they don't slow indexing or pollute results. A later lesson covers ignore files and what they do and don't protect.

How this helps your prompts: with a good index you can describe behavior instead of locations, "fix the bug where archived projects still appear in search", and Agent will find candidates. In a very large or unfamiliar repo, still point at an example file or folder when you can; semantic search narrows things down, but it can surface similar-looking code from the wrong module.