RAG service security

[cvirtucio]

I am considering enabling the RAG service, but I have some concerns about security. I know it has to mount files from the host to index files in the codebase. Will it be indexing dot files, as well? We keep a lot of dot files (.env or .envrc) that export environment variables containing secrets in our repositories. I am worried that these would get indexed and sent to the LLM.

[cvirtucio]

well, I ran a small test and the answer might be that these files are ignored, but would be great to get some confirmation:

% echo 'export THIS_IS_A_SECRET=foo' > .envrc

:lua require('avante.rag_service').add_resource('.envrc')

...

INFO:     10.88.0.7:56958 - "GET /api/v1/resources HTTP/1.1" 200 OK
INFO:     10.88.0.7:56972 - "POST /api/v1/add_resource HTTP/1.1" 422 Unprocessable Entity

hmm, I get the same result adding a regular python file

[cvirtucio]

ok. after some digging, I figured out how the RAG service works with the filesystem. it only scans files which are do not match the spec, which would be gitignored and gitcrypted files. looks fine, but the mount of the HOME directory by default struck me as a red flag. thankfully that's not a major issue since only the project root is scanned, which is translated to the filepath on the container's filesystem.

good enough for me, since the .env and .envrc files are always gitignored.