Controlling the RAG Search Paths

[inonitz]

Hi, first of all I'd like to thank you for building this extension!
I had trouble figuring out the javascript configurations, but eventually I got it to work.

For context, I'm setting up a local llm instance using your extension, and I had trouble with the RAG context.
Firstly I'm well aware of the options available currently for "filtering" the RAG Context that the extension retrieves, see the link here.

Still then, I have a project that has ~2500 Files, many of which are from git submodules,
and I don't necessarily want to retrieve all of them.
On a usual day, clangd usually consumes ~7GiB on my current project (using a simple compile_commands.json generated with CMake)

With the RAG Context enabled and clangd doing all the indexing llama-vscode wants, it can jump to as high as ~25GiB (spills to disk), which is completely unacceptable and it completely halts my workflow.

Technically Yes, I can disable the RAG, give the Chat Model individual files to think over, and call it a day.
Honestly though, it probably won't be as smart with the conclusions it gets to.
Also, nuking the RAG completely seems like an extreme option if I could just give it specific folders to scan.

Finally, are there any plans to expose a setting s.t one would be able to include/exclude specific folders in the RAG context?
Thanks.

[KingWebsites]

Yes, a simple Llama Files Include and Llama Files Exclude would be great.

After middling to good performance as a FIM (with Qwen Coder 2.5) I've been testing with RAG over the last 24 hours but I've had difficulty making it work (turns on, begins indexing, stops), but maybe that's because I only have the completion model configured, not the Embeddings or Chat?