Open austintlee opened 11 months ago
@minalsha @dbwiddis
Thanks @austintlee for tagging us to the issue. We really need to get this available to AI-Workflow soon. Can you please share an ETA for this work? Thank You
@austintlee from my understanding, as we already have rest APIs for Conversational Memory for external use, it is natural to have them in ML Client too.
@jngz-es Given that the Memory API is still behind a feature flag (still an experimental feature), do you think it is OK to put them in the (external) client? Also, if we want to move them into the ml client, do you think we can still make 2.11?
cc: @ylwu-amzn
^^ What PR #1417 does for this:
MemoryClient
object accessed from the MLClient
by mlClient.memory()
memoryClient.createConversation(new CreateConversationRequest())
; this introduces a dependency on the opensearch-ml-memory
subproject.Essentially, it's just an extremely thin wrapper around Client.execute()
.
Questions:
opensearch-ml-common
, which maybe we should do anyway.@jngz-es @ylwu-amzn @minalsha Totally happy to make any suggested changes, but let's agree on what exactly we want here
@HenryL27 , for your questions here are my thoughts.
As suggested in #1275, this client, which is currently only used by the RAG pipeline, can move to opensearch-ml-client so that it can be used by other applications.
@jngz-es @dhrubo-os do you guys think this should be merged with MachineLearningClient or are you thinking of a separate client. By the way, MachineLearningClient currently gets built as a shadow jar and it is not meant for internal use. The main thing I want to discuss here is whether or not we should have an external version of this memory client that makes it easy for applications to interact with memory.
Related: #1150