Skip to content

README_OLD.md

Latest commit

 

History

History
85 lines (63 loc) · 2.04 KB

README_OLD.md

File metadata and controls

85 lines (63 loc) · 2.04 KB

mem0

Author: yevanchen Version: 0.0.1 Type: tool

Description

mem0 is a memory management plugin that enables conversation history storage and retrieval for LLM applications.

workflow

Setup

  1. Get your API key from mem0 dashboard
  2. Install the package:
pip install mem0ai
  1. Initialize the client:
from mem0 import MemoryClient
client = MemoryClient(api_key="your-api-key")

dashboard

Memory Actions

add_memory

Stores conversation history and context for users.

messages = [
    {"role": "user", "content": "Hi, I'm Alex. I'm a vegetarian and I'm allergic to nuts."},
    {"role": "assistant", "content": "Hello Alex! I've noted your dietary preferences."}
]
client.add(messages, user_id="alex")

Backend logic:

  • Messages are stored in user-specific partitions using user_id
  • Supports conversation history and context storage
  • Handles message format validation and processing
  • Optimizes storage for efficient retrieval

retrieve_memory

Retrieves relevant conversation history based on queries.

query = "What can I cook for dinner tonight?"
memories = client.search(query, user_id="alex")

Backend logic:

  • Semantic search across user's memory partition
  • Returns relevant conversation snippets
  • Handles context ranking and relevance scoring
  • Optimizes query performance

Usage in Dify

  1. In Dify workflows, place retrieve_memory before LLM calls to provide context
  2. Add add_memory after LLM responses to store new interactions
  3. user_id can be customized in workflow run API
  4. Note: iframe and webapp modes currently don't support user_id due to lack of access control

Maybe Future Features

  • Multimodal Support
  • Memory Customization
  • Custom Categories & Instructions
  • Direct Import
  • Async Client
  • Memory Export
  • Webhooks
  • Graph Memory
  • REST API Server
  • OpenAI Compatibility
  • Custom Prompts

For feature requests or discussions, contact evanchen@dify.ai