AgentCore Memory Store limited to BaseMessages

[ltoniazzi]

In here the user is forced to store BaseMessages only.

But I feel this is rather limiting. For example if in the graph I want to cache the response from a generic API call that gives me a generic dictionary as a response, then I need to hack it into BaseMessages if my DB backend is AgentCore (while it could stay as dict for Postgres).

Is this choice related to the fact that retrieval is from memory_records (code) and those have conversational conditions on what can be stored?

Can you help me understand this design? (Also related to Issue #708)

[michaelnchin]

cc: @jgordley @bergjaak