How to recall query-relevant memory from "Conversation Variables "

[AugustLHHHHHH]

We are designing a feature within chatflow called "memory variables" to store information that the LLM needs to remember. Memory variable's scope can be set to either session-based or user-based, such as chat history, order ID, user preferences, etc. You can orchestrate the read and write logic of memory variables within chatflow, making it more flexible.I hope this helps with your issue.

Originally posted by @VincePotato in #4149

Hi,
Your blog simulats OpenAI Memory Features，where a llm node decides which memories to extract from a question, and then these memories are appended to the session variables 'memory'

how do you recall only the relevant memories before answering the question, rather than using all the memories to answer the question?

[dosubot[bot]]

To recall only the relevant memories from "Conversation Variables" in Dify, you can use the Knowledge Base Retrieval Node. This node allows you to configure retrieval strategies to determine which memories are relevant to a given query. The process involves:

1. Selecting a Query Variable: Typically, this is the user's input question.

2. Choosing a Retrieval Strategy: Options include vector search, full-text search, and hybrid search. Each strategy has its own method of determining relevance:

   • Vector Search: Converts the query into a vector and finds the closest matches in the knowledge base.
   • Full-Text Search: Uses keyword matching to retrieve relevant text fragments.
   • Hybrid Search: Combines both vector and full-text search for optimal results.

3. Configuring Retrieval Settings: You can adjust settings like TopK (number of text fragments to select) and Score Threshold (minimum similarity score) to fine-tune the retrieval process.

These strategies ensure that only the most relevant memories are appended to the session variables, enhancing the accuracy of the responses provided by the LLM ^[1][2].

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Bug Report | Other}

[Yevanchen]

In a certain memory size, you can use prompt words and assemble them all into context to ensure the effect. When the session variable exceeds the context range, there may be no good solution in the current diify.