Microsoft.SemanticKernel.Memory vs Kernel-Memory

[hemantkathuria]

I did look at https://github.com/microsoft/kernel-memory?tab=readme-ov-file#kernel-memory-km-and-sk-semantic-memory-sm

But what will be retained in future? Any roadmap on supportability for both?

Also, can kernel-memory be used to store chat history?
How do I enable detailed logging so that we can see what prompt was sent, what intent was derived, what vector embedding were generated?

Using it in serverless mode with sementic kernel

[dluc]

Hi @hemantkathuria - high level, there's a fundamental difference that KM is a service, you can deploy and use it, while SK allows to build apps and services.
That said, there are several things we learned and keep learning while we keep working on KM, and the plan is to port some features from KM to SK Memory. For instance, the ability to filter records not just by similarity, and the ability to plug in custom tables/collections, regardless of their schema.
Features that currently we don't plan to port to SK are:

• running Memory as a service
• automatically parsing files and extracting memories
• customizable asynchronous pipelines
• explicit RAG implementations
• integration with Azure AI Document Intelligence