Multiturn conversation

[HGF1KOR]

How does Mem0 solve this issue in vector similarity search?

In vector similarity search, this kind of problem can always occur:

[
  { "role": "user", "content": "Who is the Prime Minister of India?" },
  { "role": "assistant", "content": "Narendra Modi is the Prime Minister of India." },
  { "role": "user", "content": "Who is the President of the United States?" },
  { "role": "assistant", "content": "Joe Biden is the President of the United States." },
  { "role": "user", "content": "Where was he born?" }
]

In this scenario, when we would perform similarity search it would pull both the messages but the model might misunderstand who he refers to — it could incorrectly associate he with Narendra Modi instead of Joe Biden.

[Nevermetyou65]

Interesting. I want to know this too. My guess is that If we use think model it would say something like Where was he born? is lack of context and ask you to clarify

[parshvadaftari]

@HGF1KOR This conversation wouldn't be stored into the memory. This will be a conversation history. Until and unless you use custom prompt to store it then and then it will store it in mem0 as a memory. This seems to be a normal conversation about Q&A with the LLM or assistant. There's no useful information to be stored as memory.