I can’t seem to find any evidence during my agent runs that they are using semantic search (vector search, RAG, etc). I’m using the Cursor Editor similar to VSCode. The settings panel suggests my codebase is “100% indexed” however when I do runs I see no semantic search logs, just greps and file reads, and when I ask the agent (Opus 4.8 or Composer 2.5) if they have semantic search, they say they don’t and list out the tools they do have.
So I’m confused, do I have to do something to enable semantic search? Is it automatically disabled for my codebase (it’s 250 files)? Why isn’t it available?
Hey, thanks for the request. Everything looks fine. You haven’t misconfigured anything, and there’s nothing you need to enable separately.
The set of tools the agent gets is defined on our side and changes over time. Right now the design agent mainly relies on grep plus file reads to navigate the code, especially newer models like Composer 2.5 that are tuned for that kind of search. So seeing only greps and file reads in the logs is expected behavior, not a regression.
A couple notes:
Don’t rely on what the model says about its own tools. Models often list the wrong tools, so it’s not a reliable way to check.
100% indexed isn’t wasted. The codebase index still helps with retrieval. For a repo with around 250 files, grep plus reads is often faster and more accurate anyway.
If you notice the agent can’t find the right context and answer quality drops, let me know and we’ll take a closer look.
P.S. I see this request was also posted in a second thread. Let’s continue here, and I’ll close the other one as a duplicate so the discussion doesn’t split.