Kernel Memory without OpenAI

[arafattehsin]

Hi Team

I couldn't find any answer related to this so thought to post it here. I have been able to load the model locally with local embeddings. This means that I am not leveraging internet to do text-based RAG and it is working just fine.

I have written down a blog post on how I have achieved it using SemanticTextMemory

Now I want to add documents to it and I am aware that SemanticTextMemory can't handle it. Therefore, I am looking for other options and Kernel Memory is one of the primary candidate in that list. However, after exploring it, I have found out that it is tightly coupled with OpenAI services (for embeddings). Is there a way we can use them without OpenAI service and leverage whatever I am using in my existing application.

P.S: If you have not read my blog post, I am using bge-micro-v2 for local embeddings and ITextEmbeddingGenerationService for text generation.

Thanks.

[sophialagerkranspandey]

We are wondering if you are using Semantic Memory or kernel memory here (https://github.com/microsoft/kernel-memory)?

[arafattehsin]

My apologies @sophialagerkranspandey, I am using Kernel Memory but I just realised that I should have posted in their discussion forum :) I will close this discussion.