云原生数据仓库AnalyticDB：AnalyticDB for MySQL + Mem0 实践指南

准备 AnalyticDB for MySQL

1. 在您的 AnalyticDB for MySQL 集群中创建一个用于本教程的数据库。

   CREATE DATABASE mem0;

2. 如果您计划通过公网连接，请在 AnalyticDB for MySQL 控制台为集群开启公网访问地址，并将您运行 Demo 环境的公网 IP 地址添加到白名单中。

安装 Mem0 及依赖

1. 检查 Python 环境：确保您的运行环境中已安装 Python。本教程在 Python 3.12.2 版本下验证通过。

   # 使用Mac, 检查Python版本, 如无请至官网安装 https://www.python.org/
   python3 --version
   
   # 使用Linux, 检查Python版本, 如无使用 yum 进行安装, sudo yum install python
   python --version

2. 安装 Mem0 框架：解压您获取的 Mem0 框架文件，并安装必要的依赖库。

   # 解压并安装
   unzip mem0.zip 
   cd mem0
   pip install .
   pip install pymysql dbutils

配置环境变量：

为了安全起见，建议将大模型的 API Key 配置为环境变量。

echo "export DASHSCOPE_API_KEY='<YOUR_DASHSCOPE_API_KEY>'" >> ~/.bashrc
# 可选,默认为新加坡
echo "export DASHSCOPE_BASE_URL='https://dashscope-intl.aliyuncs.com/compatible-mode/v1'" >> ~/.bashrc   

source ~/.bashrc

#检查是否生效
echo $DASHSCOPE_API_KEY
echo $DASHSCOPE_BASE_URL  

初始化 Mem0 实例：

在您的 Python 脚本中，配置并初始化一个 Memory 实例。

• LLM 配置：使用阿里云百炼的 qwen-plus 模型。

• Embedder 配置：使用阿里云百炼的 text-embedding-v4 模型，向量维度为 1536。

• Vector Store 配置：使用 aliyun_adb，并填入您的 AnalyticDB for MySQL 连接信息。

from mem0 import Memory

# This example must connect to an adb instance with 1536 vector dimensions specified.
config = {
    "llm": {
        "provider": "bailian",
        "config": {"model": "qwen-plus"},
    },
    "embedder": {
        "provider": "bailian",
        "config": {"model": "text-embedding-v4", "embedding_dims": 1536},
    },
    "vector_store": {
        "provider": "aliyun_adb",
        "config": {
            "host": "<adb host>",
            "port": 3306,
            "database": "mem0",
            "user": "<your user name>",
            "password": "<your password>",
            "collection_name": "adb_mem0_demo"
        }
    }
}


"""
Suppose we have two users, Mike and Julia
1、Mike is a vegetarian, who likes fruit and playing basketball
2、Julia is a carnivore, who likes steak and playing badminton
"""

m = Memory.from_config(config_dict=config)

添加与获取记忆：

使用 m.add() 方法为用户添加记忆，m.get_all() 方法获取用户的全部记忆。

• 代码

  # add memorys for Mike
  result = m.add("I'm a vegetarian", user_id="Mike")
  memory_id = result["results"][0]["id"] ,
  print("add a memory for Mike:" + "I'm a vegetarian" + ", result => ", result , "\n")
  
  
  result = m.add("I like durian", user_id="Mike")
  print("add a memory for Mike:" + "I like durian" + ", result => " , result , "\n")
  
  # Get all memories for a user
  result = m.get_all(user_id="Mike")
  print("Get all memories for a user:" + "Mike" + ", result => " , result , "\n")

• 结果分析
  返回的结果是一个列表，包含了 Mike 的两条记忆及其元数据，如 ID、内容、创建时间等。

  add a memory for Mike:I'm a vegetarian, result =>  {'results': [{'id': '32ac2932-10da-422b-beba-bb5a2b4666aa', 'memory': "I'm a vegetarian", 'event': 'ADD'}]} 
  
  add a memory for Mike:I like durian, result =>  {'results': [{'id': '40098c00-9f96-4486-af8b-5c79612fea10', 'memory': 'I like durian', 'event': 'ADD'}]} 
  
  Get all memories for a user:Mike, result =>  {'results': [{'id': '40098c00-9f96-4486-af8b-5c79612fea10', 'memory': 'I like durian', 'hash': 'dfd43e3d0a6ddc82cda440e0a1d39e7f', 'metadata': None, 'created_at': '2025-10-29T20:52:54.156926-07:00', 'updated_at': None, 'user_id': 'Mike'}, {'id': '32ac2932-10da-422b-beba-bb5a2b4666aa', 'memory': "I'm a vegetarian", 'hash': '27f3478f622dd27dd71969f131bd886f', 'metadata': None, 'created_at': '2025-10-29T20:52:51.126628-07:00', 'updated_at': None, 'user_id': 'Mike'}]} 

更新与删除记忆：

Mem0 支持通过记忆 ID 手动删除，或在添加新记忆时自动处理冲突。

• 代码

  result = m.delete(memory_id=memory_id)
  print("Manual Delete a memory:" + "I like durian" + ", result => " , result , "\n")
  
  result = m.add("I hate things that taste like durian", user_id="Mike")
  print("Automatic Delete a contradictory historical memory for Mike:" + "I like durian" + ", result => " , result , "\n")
  
  # Get all memories for a user
  result = m.get_all(user_id="Mike")
  print("Get all memories for a user:" + "Mike" + ", result => " , result , "\n")

• 结果分析

  Manual Delete a memory:I like durian, result =>  {'message': 'Memory deleted successfully!'} 
  
  Automatic Delete a contradictory historical memory for Mike:I like durian, result =>  {'results': [{'id': 'dc308443-ae8c-4cb3-851f-060adfc39eaf', 'memory': 'I like durian', 'event': 'DELETE'}]} 
  
  Get all memories for a user:Mike, result =>  {'results': []} 

  • 添加冲突记忆后，add 方法的返回结果会包含一个 DELETE 事件，表明旧的、冲突的记忆（“I like durian”）已被自动移除。

  • 再次调用 get_all 时，会发现旧的记忆已不复存在，体现了 Mem0 的自适应与自洽能力。

检索与搜索记忆：

Mem0 支持通过元数据进行精确过滤检索，以及通过语义进行相似性搜索。

1. 准备数据：为 Mike 和 Julia 添加带有 agent_id 和 metadata 的记忆。

   • 代码

     # Mike Memory
     result = m.add("I'm a vegetarian", user_id="Mike", agent_id="diet-assistant")
     print("add a memory for Mike:" + "I'm a vegetarian" + ", result => ", result , "\n")
     
     result = m.add("I like durian, watermelon and strawberries", user_id="Mike", metadata={"category": "hobbies"})
     print("add a memory for Mike with meta:" + "I like durian, watermelon and strawberries" + ", result => " , result , "\n")
     
     result = m.add("I like playing basketball", user_id="Mike", agent_id="fitness-assistant")
     print("add a memory for Mike:" + "I like playing basketball" + ", result => " , result , "\n")
     
     
     # Julia Memory
     result = m.add("I'm a carnivore", user_id="Julia", agent_id="diet-assistant")
     print("add a memory for Julia:" + "I'm a carnivore" + ", result => " , result , "\n")
     
     result = m.add("I like steak", user_id="Julia", metadata={"category": "hobbies"})
     print("add a memory for Julia:" + "I like steak" + ", result => " , result , "\n")
     
     result = m.add("I like playing badminton", user_id="Julia", agent_id="fitness-assistant")
     print("add a memory for Julia:" + "I like playing badminton" + ", result => " , result , "\n")
     

   • 结果分析

     add a memory for Mike:I'm a vegetarian, result =>  {'results': [{'id': 'a0fadb7e-c6b7-46b2-b0b8-6b036d879015', 'memory': "I'm a vegetarian", 'event': 'ADD'}]} 
     
     add a memory for Mike with meta:I like durian, watermelon and strawberries, result =>  {'results': [{'id': 'efec6c25-c2a7-454f-8344-367018c5b607', 'memory': 'I like durian, watermelon and strawberries', 'event': 'ADD'}]} 
     
     add a memory for Mike:I like playing basketball, result =>  {'results': [{'id': '75da451c-dbf7-45aa-b9ab-72aab54df7b8', 'memory': 'I like playing basketball', 'event': 'ADD'}]} 
     
     add a memory for Julia:I'm a carnivore, result =>  {'results': [{'id': '93a03639-195e-40ef-9f5f-55d366b76333', 'memory': "I'm a carnivore", 'event': 'ADD'}]} 
     
     add a memory for Julia:I like steak, result =>  {'results': [{'id': 'e30faceb-ba71-4b27-b052-28863f26abf9', 'memory': 'I like steak', 'event': 'ADD'}]} 
     
     add a memory for Julia:I like playing badminton, result =>  {'results': [{'id': '90d45867-b198-415f-99e8-7375b4c6c35b', 'memory': 'I like playing badminton', 'event': 'ADD'}]} 

2. 精确过滤检索：使用 get_all 方法并传入过滤条件。

   • 代码

     # Get all memories for a specific agent belonging to a user
     result = m.get_all(user_id="Mike", agent_id="fitness-assistant")
     print("Get all memories for a specific agent belonging to a user:" + "Mike" + ", result => " , result , "\n")
     
     
     # Get all memories belonging to a user with filter
     result = m.get_all(user_id="Julia", filters={"category": "hobbies"})
     print("Get all memories belonging to a user with filter:" + "Julia" + ", result => " , result , "\n")

   • 结果分析

     Get all memories for a specific agent belonging to a user:Mike, result =>  {'results': [{'id': '75da451c-dbf7-45aa-b9ab-72aab54df7b8', 'memory': 'I like playing basketball', 'hash': '000b24fa8e8f76134286d2c4626c036d', 'metadata': None, 'created_at': '2025-10-30T00:13:08.390949-07:00', 'updated_at': None, 'user_id': 'Mike', 'agent_id': 'fitness-assistant'}]} 
     
     Get all memories belonging to a user with filter:Julia, result =>  {'results': [{'id': 'e30faceb-ba71-4b27-b052-28863f26abf9', 'memory': 'I like steak', 'hash': '6e7129fc2010e93e94c943dbf6f69246', 'metadata': {'category': 'hobbies'}, 'created_at': '2025-10-30T00:13:14.631531-07:00', 'updated_at': None, 'user_id': 'Julia'}]} 

3. 相似性搜索：使用 search 方法进行语义搜索。返回结果按相似度得分（score）降序排列。

   • 代码

     # Search memories for a user
     result = m.search("Who likes playing badminton better", agent_id="fitness-assistant")
     print("Search memories:" + "Who likes playing badminton better" + ", result => " , result , "\n")
     
     # Search memories for a specific agent belonging to a user
     result = m.search("Who will celebrate on International Vegetarian Day", agent_id="diet-assistant")
     print("Search memories:" + "Who will celebrate on International Vegetarian Day" + ", result => " , result , "\n")
     
     # Search memories for a user
     result = m.search("what kind of fruit do you like", user_id="Mike")
     print("Search memories:" + "what kind of fruit do you like." + ", result => " , result , "\n")
     
     # Search memories for a user
     result = m.search("what sport do you like", user_id="Julia", agent_id="fitness-assistant")
     print("Search memories:" + "what sport do you like." + ", result => " , result , "\n")
     

   • 结果分析

     Search memories:Who likes playing badminton better, result =>  {'results': [{'id': '90d45867-b198-415f-99e8-7375b4c6c35b', 'memory': 'I like playing badminton', 'hash': '304d03421834255d1650b46877b3785b', 'metadata': None, 'score': 0.82532454, 'created_at': '2025-10-30T00:13:18.218251-07:00', 'updated_at': None, 'user_id': 'Julia', 'agent_id': 'fitness-assistant'}, {'id': '75da451c-dbf7-45aa-b9ab-72aab54df7b8', 'memory': 'I like playing basketball', 'hash': '000b24fa8e8f76134286d2c4626c036d', 'metadata': None, 'score': 0.56634325, 'created_at': '2025-10-30T00:13:08.390949-07:00', 'updated_at': None, 'user_id': 'Mike', 'agent_id': 'fitness-assistant'}]} 
     
     Search memories:Who will celebrate on International Vegetarian Day, result =>  {'results': [{'id': 'a0fadb7e-c6b7-46b2-b0b8-6b036d879015', 'memory': "I'm a vegetarian", 'hash': '27f3478f622dd27dd71969f131bd886f', 'metadata': None, 'score': 0.5640734, 'created_at': '2025-10-30T00:13:01.011810-07:00', 'updated_at': None, 'user_id': 'Mike', 'agent_id': 'diet-assistant'}, {'id': '93a03639-195e-40ef-9f5f-55d366b76333', 'memory': "I'm a carnivore", 'hash': '5f81d3e6f06baba2e10ec6c155d012e3', 'metadata': None, 'score': 0.41749594, 'created_at': '2025-10-30T00:13:11.462159-07:00', 'updated_at': None, 'user_id': 'Julia', 'agent_id': 'diet-assistant'}]} 
     
     Search memories:what kind of fruit do you like., result =>  {'results': [{'id': 'efec6c25-c2a7-454f-8344-367018c5b607', 'memory': 'I like durian, watermelon and strawberries', 'hash': '07cc049e2beaa043bb359b50a33664ac', 'metadata': {'category': 'hobbies'}, 'score': 0.6519712, 'created_at': '2025-10-30T00:13:04.981195-07:00', 'updated_at': None, 'user_id': 'Mike'}, {'id': '75da451c-dbf7-45aa-b9ab-72aab54df7b8', 'memory': 'I like playing basketball', 'hash': '000b24fa8e8f76134286d2c4626c036d', 'metadata': None, 'score': 0.3975049, 'created_at': '2025-10-30T00:13:08.390949-07:00', 'updated_at': None, 'user_id': 'Mike', 'agent_id': 'fitness-assistant'}, {'id': 'a0fadb7e-c6b7-46b2-b0b8-6b036d879015', 'memory': "I'm a vegetarian", 'hash': '27f3478f622dd27dd71969f131bd886f', 'metadata': None, 'score': 0.3382097, 'created_at': '2025-10-30T00:13:01.011810-07:00', 'updated_at': None, 'user_id': 'Mike', 'agent_id': 'diet-assistant'}]} 
     
     Search memories:what sport do you like., result =>  {'results': [{'id': '90d45867-b198-415f-99e8-7375b4c6c35b', 'memory': 'I like playing badminton', 'hash': '304d03421834255d1650b46877b3785b', 'metadata': None, 'score': 0.61615115, 'created_at': '2025-10-30T00:13:18.218251-07:00', 'updated_at': None, 'user_id': 'Julia', 'agent_id': 'fitness-assistant'}]} 

     • 搜索羽毛球：agent_id 过滤了非运动相关的记忆，结果中“I like playing badminton”的得分最高，排在首位。

     • 搜索素食日：agent_id 过滤了非饮食相关的记忆，结果中“I'm a vegetarian”的得分最高，排在首位。

     • 搜索 Mike 的水果偏好：user_id 过滤了 Julia 的记忆，结果中“I like durian, watermelon and strawberries”得分最高。

     • 搜索 Julia 的运动偏好：user_id 和 agent_id 双重过滤，精确地返回了“I like playing badminton”。

清理记忆：

使用 delete_all 方法可以清空指定用户或 Agent 的所有记忆。

• 代码

  # Delete all memories for a specific agent belonging to a user
  m.delete_all(user_id="Julia")
  result = m.get_all(user_id="Julia")
  print(result)
  
  # Delete all memories for a user
  m.delete_all(user_id="Mike")
  result = m.get_all(user_id="Mike")
  print(result)

• 结果分析

  Resetting collection mem0_adb_demo...
  {'results': []}
  Resetting collection mem0_adb_demo...
  {'results': []}