Is there any memory update strategy exist in memobase: avoiding the same content repeatedly inserted

[Cheeeeyan]

I have deployed the Memobase service locally using Docker. Interestingly, I noticed in the user_profilesof the Memobase vector database that the summaries of user information are duplicated. It performs like:

{
    "content": "Xiao Ge; Xiao Ge; Xiao Ge; Xiao Ge; Xiao Ge",
    "created_at": "2025-11-04T03:05:29.140016Z",
    "id": "1aa64d70-7feb-477c-afdd-a716aaeb3e5e",
    "sub_topic": "name",
    "topic": "basic_info",
    "updated_at": "2025-11-13T05:07:59.808905Z"
},
{
    "content": "Likes hamburgers and steak; Likes hamburgers and steak",
    "created_at": "2025-11-04T03:05:29.140016Z",
    "id": "c34c9ee0-1048-47be-b772-6f3b3457b7a0",
    "sub_topic": "food",
    "topic": "hobbies",
    "updated_at": "2025-11-05T05:52:02.367032Z"
}

I am using the memobase plugin in Dify. My workflow follows the official usage, where I only set up the system to insert data into Memobase simultaneously when the LLM replies. However, it seems like the information from each conversation is being repeatedly updated into the Memobase vector database without considering duplicate content. Does Memobase have an automatic mechanism to judge information and avoid storing identical data in the database?

[gusye1234]

Which LLM model you're using?

We let the llm to update the profile slot based on the new-coming memory, so if the LLM is unable to understand the prompt, it may performs badly.

[0000005]

What model do you recommend using? I’ve also run into the issue of the profile not updating. I tried glm-4.5 and glm4.5-air, and the results haven’t been very good.