Hey, thanks for the report. This isn’t a bug. The Codebase Indexing section, including the Delete Index button, was removed on purpose. We’re phasing out semantic indexing in favor of Instant Grep. Models now work great by grepping directly through files, so a separate semantic index isn’t needed anymore.
Key point about this change. Cursor no longer builds a semantic index of your code and it doesn’t compute or store embeddings on servers. Search runs locally over your files, and .cursorignore and .gitignore are still respected. So there’s basically nothing to “delete,” which is why the button is gone.
What was your goal with Delete Index? Were you trying to free up disk space, reset state, or remove previously uploaded data? I can tell you what to do next.
I was investigating high token usage and i thought one potential culprit was some big sql files in my project.
So I added them in the .cursorignore file and then checked the “view included files” in “Cursor settings > Indexing & docs” just to be sure the files were removed.
Since the files were still there I thought I had to refresh the index.
Is the embeddable_files.txt file still used or is it going to be removed?
Got it. There are two separate things here, so let me break it down.
View included files and embeddable_files.txt are part of the old semantic indexing pipeline that we’re currently phasing out. That’s why the button and the whole section are gone. There’s no separate index or embeddings anymore, so there’s nothing to refresh. I wouldn’t rely on that list since it’s part of the retiring system and may show an outdated state.
For your goal, excluding large SQL files so they don’t eat up context, the right tool now is .cursorignore. It blocks files from both indexing and Agent access, so the model won’t pull them into context. .gitignore is respected too. More details here: Ignore files | Cursor Docs
A couple notes:
Patterns in .cursorignore work like gitignore, for example *.sql or path/to/dumps/.
Terminal commands and MCP tools run outside these limits, so they can still read those files, that’s expected.
Let me know if token usage dropped after adding the patterns. If not, we can look at what’s still getting pulled into context.
