Require 'context-dependent interpretation' feature while storing actual content in the memmory

[hariumeshuk]

Hello,
I came across a situation where mem0 is lacks in storing actual result context in the memory.
Let me take the real example i used
Below is the insertion code i used

## Actual result_content retrieved by the model after combining sql result against the user query
result_content = result_str [
  {
    "response": " The cheapest product we have is the 'Facial Cleanser Foam'. This product is designed to gently cleanse your skin, removing dirt and impurities while maintaining moisture balance. It's great for daily use and suitable for various skin types. If you're looking for an affordable option to keep your skin fresh and clean, this facial cleanser could be a perfect choice."
  }
]
memory.add(messages=result_content, user_id=id, infer=True)

The above result content was retrieved by the model combining with sql result-
eg:

user_query = "which is the cheapest product you have"

however the mem0 stored below context in the memory

### mem0  result
"Facial Cleanser Foam is designed to gently cleanse skin, removing dirt and impurities while maintaining moisture balance, suitable for daily use and various skin types, and is a perfect affordable option to keep skin fresh and clean"

Here mem0 didn't evaluate what was the context that user must have asked and even the 'actual result' contains user's real ask - the sentence: 'The cheapest product we have is the 'Facial Cleanser Foam' which is missed to store in the memory.

This means, mem0 lacks in storing actual context, rather it prompting to store generic context.

From a user perspective who interact with conversational AI application, a user expect to see the answer
with context 'The cheapest product we have is the 'Facial Cleanser Foam' and then - semantic context that mem0 used.

Because the reasoning agent refer the mem0 for the same/similar question rather querying sql db- lets say- a caching mechanism.

[aniruddhaadak80]

From my point of view, the problem is that mem0 preserved descriptive product information but dropped the relational fact that actually answered the user question, which in this case is cheapest.

That means the stored memory is semantically rich but operationally weak. If a later query asks for the cheapest item again, the system may retrieve a good description of the product without preserving why that product was selected in the first place.

I think this needs either structured fact extraction alongside the summary, or a memory type that explicitly keeps comparative and query bound conclusions with provenance. Otherwise the cache becomes less useful for exactly the cases where users expect it to help.