Anyone using a "decoupled" LibreChat instance if it can be done (Pinecone for RAG, MongoDB Atlas, etc)?

[danielrosehill]

Hello LibreChat users!

I've been doing a bit of tweaking with my instance lately as I set it up to experiment with it and never expected to actually find myself using it all the time. Testament to a great tool!

I'm trying to think about scalability from the outset. My LibreChat instance itself is on a pretty non-remarkable VPS that I use for a couple of AI projects. I'm thinking now about how to set it up a bit more robustly and also begin thinking about a proper backup strategy to better protect the data.

To re-architect the deployment for scalability, I thought that a good place to start would be decoupling some of the components ... but that requires understanding what's doing what in the tool.

For example my first inclination was to see if I could configure an S3 bucket to write all the files into ... but if all the uploads are ultimately going into the vector store, are regular file storage buckets even applicable here?

Pinecone seems like a great option for vector storage and I see that it's supported. For MongoDB, I see that you can use Atlas instead of the local one.

I'm wondering if anyone has actually deployed a instance along these lines? Or perhaps rolled out a LibreChat instance on something like GCP, AWS, etc. where it's easy to create the modules and linked them up?

Finally, has anyone attempted to migrate data between two instances? It would be nice to take my time setting up the new one and then try to roll in the data if that were possible. But it may be one of those situations where it's easier just to start from scratch.

Either way, any perspectives appreciated!

TIA!

[fuegovic]

To migrate the data between mongo databases, you can use mongodump & mongorestore
I also successfully used Studio 3T in the past to backup/move/restore the MongoDB database

[danielrosehill]

@fuegovic Thx!

[mkagit]

I use Atlas for main DB and RAG. Still waiting for AWS S3 integration.