RAG_EMBEDDING_MODEL location on offline mode with ollama backend?

[ArKam]

Hi everyone,

I've few question regarding RAG and OpenWebUI in OFFLINE context.

If I'm not making any mistake, when we set the RAG_EMBEDDING_MODEL envvar with let say sentence-transformers/all-MiniLM-L6-v2 OpenWebUI is looking for: /app/backend/data/cache/embedding/models/

To my understanding, it means OpenWebUI will run the model itself without leveraging any backend model engine right?

Is there a way to let OpenWebUI delegate this model run to our backend (ollama currently)?
If so, how can I set OpenWebUI to do so? I mean, do I just need to load ollama with the model?

Currently our OpenWebUI container is on a host that doesn't have any GPU but it is using our ollama backend deployment which itself IS hosted on GPUs based hosts.

Both zones doesn't have access to the internet at all, but we load models on ollama on our own.
Right now, we did loaded Qwen3:8B / Qwen3-embedding:8B and Qwen3-reranker:8B sucessfully, but we would be sure OpenWebUI can indeed use the embedding and reranker models from the ollama instances.

Thanks everyone!

[ArKam]

Arff... nevermind, seems like I just missed the RAG_EMBEDDING_MODEL envvars from the documentation...

[rgaricano]

do you mean RAG_EMBEDDING_ENGINE instead?