RAG API Vector Storage

[JaipuriaAI]

I have deployed LibreChat On Railway using MongoDB.
The RAG API is working fine, but i am unable to see the vector data of these files. Please guide me as to where i can see this data.
I want to set a CRON job as well, that will delete files after 30 days.

[akshayi1]

I use MongoDB Atlas and I can see the files and indexes just fine on the UI there.

[JaipuriaAI]

Thanks! I can see the files in MongoDB Atlas, but I'm specifically looking for the vector embeddings data. Are you seeing the actual vector arrays in the collections?

[onestardao]

Good question — I’ve run into similar confusion with vector storage visibility in RAG setups.

In most pipelines, especially ones using MongoDB as backing store, the vectors themselves often get encoded or abstracted through an additional layer (like external embedding services or separate indexing layers), so they may not appear directly inside your collections unless you’re explicitly persisting them.

That “invisible vector” issue tends to trip up monitoring and debugging, especially when trying to validate similarity matches. It’s one of the silent failure patterns we ended up documenting and building tooling around — semantic gaps that don’t show up in logs.

If you're building anything production-grade, I’d strongly recommend treating embedding storage visibility and memory layout as part of your infra checks. Happy to share more if you’re digging into this area — we’ve published some MIT-licensed tools and frameworks backed by tesseract.js folks, but I won’t drop links unless asked.