Mem0 provides two core operations for managing memories in AI applications: adding new memories and searching existing ones. This guide covers how these operations work and how to use them effectively in your application.

​
Core Operations

Mem0 exposes two main endpoints for interacting with memories:

  • The add endpoint for ingesting conversations and storing them as memories
  • The search endpoint for retrieving relevant memories based on queries

​
Adding Memories

Architecture diagram illustrating the process of adding memories.

The add operation processes conversations through several steps:

  1. Information Extraction

    • An LLM extracts relevant memories from the conversation
    • It identifies important entities and their relationships

  2. Conflict Resolution

    • The system compares new information with existing data
    • It identifies and resolves any contradictions

  3. Memory Storage

    • Vector database stores the actual memories
    • Graph database maintains relationship information
    • Information is continuously updated with each interaction

​
Searching Memories

Architecture diagram illustrating the memory search process.

The search operation retrieves memories through a multi-step process:

  1. Query Processing

    • LLM processes and optimizes the search query
    • System prepares filters for targeted search

  2. Vector Search

    • Performs semantic search using the optimized query
    • Ranks results by relevance to the query
    • Applies specified filters (user, agent, metadata, etc.)

  3. Result Processing

    • Combines and ranks the search results
    • Returns memories with relevance scores
    • Includes associated metadata and timestamps

This semantic search approach ensures accurate memory retrieval, whether you’re looking for specific information or exploring related concepts.