Scenario 9: User Profile Management

This scenario demonstrates powermem's UserMemory feature - automatic user profile extraction, management, and integration with memory search.

Prerequisites

Python 3.10+
powermem installed (pip install powermem)
LLM provider configured (for profile extraction)
OceanBase configured as vector store (UserMemory requires OceanBase)

Important Note: UserMemory requires OceanBase as the storage backend. If you attempt to use UserMemory with a different storage provider, it will raise a ValueError. Please configure OceanBase as your vector store provider.

Understanding User Profile Management

UserMemory extends Memory with user profile capabilities:

Automatic profile extraction: Extracts user-related information from conversations (name, profession, interests, preferences, etc.)
Continuous profile updates: Updates and refines the profile based on new conversations
Profile storage: Stores profiles separately from memories for efficient retrieval
Joint search: Optionally includes profile information when searching memories

Step 1: Initialize UserMemory

First, let's initialize UserMemory with proper configuration:

# user_profile_example.py
from powermem import UserMemory, auto_config

# Load configuration (auto-loads from .env or uses defaults)
config = auto_config()

# Create UserMemory instance
user_memory = UserMemory(config=config)

print("✓ UserMemory initialized successfully!")

Run this code:

python user_profile_example.py

Expected output:

✓ UserMemory initialized successfully!

Step 2: Add Conversation and Extract Profile

Add a conversation and automatically extract user profile information:

# user_profile_example.py
from powermem import UserMemory, auto_config

config = auto_config()
user_memory = UserMemory(config=config)

# Add a conversation
conversation = [
    {"role": "user", "content": "Hi, I'm Alice. I'm a 28-year-old software engineer from San Francisco. I work at a tech startup and love Python programming."},
    {"role": "assistant", "content": "Nice to meet you, Alice! Python is a great language for software engineering."}
]

result = user_memory.add(
    messages=conversation,
    user_id="user_001",
    agent_id="assistant_agent",
    run_id="session_001"
)

print(f"✓ Conversation added successfully")
print(f"  - Profile extracted: {result.get('profile_extracted', False)}")
if result.get('profile_content'):
    print(f"  - Profile content: {result['profile_content']}")
print(f"  - Memory results count: {len(result.get('results', []))}")

Run this code:

python user_profile_example.py

Expected output:

✓ Conversation added successfully
  - Profile extracted: True
  - Profile content: Name: Alice. Age: 28. Location: San Francisco. Profession: Software engineer at a tech startup. Interests: Python programming.
  - Memory results count: 1

Step 3: Update Profile with More Conversations

Add more conversations to update and refine the user profile:

# user_profile_example.py
from powermem import UserMemory, auto_config

config = auto_config()
user_memory = UserMemory(config=config)

# First conversation
conversation1 = [
    {"role": "user", "content": "Hi, I'm Alice. I'm a software engineer from San Francisco."},
    {"role": "assistant", "content": "Nice to meet you, Alice!"}
]

result1 = user_memory.add(
    messages=conversation1,
    user_id="user_001",
    agent_id="assistant_agent"
)

print("=== First conversation ===")
print(f"Profile extracted: {result1.get('profile_extracted', False)}")
if result1.get('profile_content'):
    print(f"Profile: {result1['profile_content']}")

# Second conversation - adds more information
conversation2 = [
    {"role": "user", "content": "I also enjoy reading science fiction novels and playing tennis on weekends."},
    {"role": "assistant", "content": "That sounds like great hobbies!"}
]

result2 = user_memory.add(
    messages=conversation2,
    user_id="user_001",
    agent_id="assistant_agent"
)

print("\n=== Second conversation ===")
print(f"Profile updated: {result2.get('profile_extracted', False)}")
if result2.get('profile_content'):
    print(f"Updated profile: {result2['profile_content']}")

Run this code:

python user_profile_example.py

Expected output:

=== First conversation ===
Profile extracted: True
Profile: Name: Alice. Profession: Software engineer. Location: San Francisco.

=== Second conversation ===
Profile updated: True
Updated profile: Name: Alice. Profession: Software engineer. Location: San Francisco. Hobbies: Reading science fiction novels, playing tennis.

Step 4: Get User Profile

Retrieve the user profile directly:

# user_profile_example.py
from powermem import UserMemory, auto_config

config = auto_config()
user_memory = UserMemory(config=config)

# Add some conversations first
conversation = [
    {"role": "user", "content": "I'm Bob, a data scientist. I love machine learning and hiking."},
    {"role": "assistant", "content": "Great to meet you, Bob!"}
]

user_memory.add(
    messages=conversation,
    user_id="user_002",
    agent_id="assistant_agent"
)

# Get the user profile
profile = user_memory.profile(
    user_id="user_002"
)

if profile:
    print("✓ Profile retrieved successfully")
    print(f"  - Profile ID: {profile.get('id')}")
    print(f"  - User ID: {profile.get('user_id')}")
    if profile.get('profile_content'):
        print(f"  - Profile content: {profile.get('profile_content', '')}")
    if profile.get('topics'):
        print(f"  - Topics: {profile.get('topics')}")
    print(f"  - Created at: {profile.get('created_at')}")
    print(f"  - Updated at: {profile.get('updated_at')}")
else:
    print("✗ No profile found")

Run this code:

python user_profile_example.py

Expected output:

✓ Profile retrieved successfully
  - Profile ID: 1234567890123456789
  - User ID: user_002
  - Agent ID: assistant_agent
  - Run ID: 
  - Profile content: Name: Bob. Profession: Data scientist. Interests: Machine learning, hiking.
  - Created at: 2024-01-15T10:30:00
  - Updated at: 2024-01-15T10:30:00

Step 5: Search Memories Without Profile

Search memories without including the profile:

# user_profile_example.py
from powermem import UserMemory, auto_config

config = auto_config()
user_memory = UserMemory(config=config)

user_id = "user_003"

# Add some conversations
user_memory.add(
    messages=[
        {"role": "user", "content": "I'm Charlie, a product manager. I work on mobile apps."},
        {"role": "assistant", "content": "Interesting!"}
    ],
    user_id=user_id,
    agent_id="assistant_agent"
)

user_memory.add(
    messages=[
        {"role": "user", "content": "I prefer agile development methodology."},
        {"role": "assistant", "content": "That's a popular approach."}
    ],
    user_id=user_id,
    agent_id="assistant_agent"
)

# Search without profile
results = user_memory.search(
    query="work and preferences",
    user_id=user_id,
    agent_id="assistant_agent",
    limit=5,
    add_profile=False  # Don't include profile
)

print(f"✓ Search completed")
print(f"  - Results count: {len(results.get('results', []))}")
print(f"  - Profile included: {'profile_content' in results}")

for i, result in enumerate(results.get('results', []), 1):
    print(f"  {i}. {result.get('memory', '')} (score: {result.get('score', 0):.2f})")

Run this code:

python user_profile_example.py

Expected output:

✓ Search completed
  - Results count: 2
  - Profile included: False
  1. Works on mobile apps as a product manager (score: 0.85)
  2. Prefers agile development methodology (score: 0.78)

Optional: Query Rewrite with User Profile

UserMemory.search() can optionally rewrite the query using the user's profile to improve recall. This feature is optional and disabled by default. It only runs when query_rewrite.enabled=True and a user_id with profile_content is available. If the profile is missing, the query is too short, or the rewrite fails, it falls back to the original query.

Practical example (profile + ambiguous query):

# user_profile_example.py
from powermem import UserMemory, auto_config

config = auto_config()
user_memory = UserMemory(config=config)

user_id = "user_rewrite_demo"

# Build profile with relocation info
user_memory.add(
    messages="Last month I moved from Chengdu to Hangzhou, and started a new job.",
    user_id=user_id,
    profile_type="content"
)

# Ambiguous query that relies on profile context
results = user_memory.search(
    query="Recommend some delicious food near my place.",
    user_id=user_id,
    limit=5
)

Contrast: rewrite enabled vs disabled

import os

# Enable query rewrite via env
os.environ["QUERY_REWRITE_ENABLED"] = "true"
user_memory_rewrite = UserMemory(config=auto_config())

# Disable query rewrite via env
os.environ["QUERY_REWRITE_ENABLED"] = "false"
user_memory_no_rewrite = UserMemory(config=auto_config())

# Same query, same user profile
query = "Recommend some delicious food near my place."

result_with_rewrite = user_memory_rewrite.search(
    query=query,
    user_id=user_id,
    limit=5
)

result_without_rewrite = user_memory_no_rewrite.search(
    query=query,
    user_id=user_id,
    limit=5
)

Example config:

config = {
    # ... other config
    "query_rewrite": {
        "enabled": True,
        # "prompt": "Rewrite queries to be specific and grounded in the user profile."
    }
}

Environment variables are also supported (see .env.example):

QUERY_REWRITE_ENABLED=false
# QUERY_REWRITE_PROMPT=
# QUERY_REWRITE_MODEL_OVERRIDE=

QUERY_REWRITE_PROMPT is optional custom instructions for rewrite.
QUERY_REWRITE_MODEL_OVERRIDE is optional and must be another model from the same LLM provider family.

Step 6: Search Memories With Profile

Search memories and include the user profile in results:

# user_profile_example.py
from powermem import UserMemory, auto_config

config = auto_config()
user_memory = UserMemory(config=config)

user_id = "user_004"

# Add conversations
user_memory.add(
    messages=[
        {"role": "user", "content": "I'm Diana, a UX designer. I love creating beautiful interfaces."},
        {"role": "assistant", "content": "That's wonderful!"}
    ],
    user_id=user_id,
    agent_id="assistant_agent"
)

# Search with profile
results = user_memory.search(
    query="user background and interests",
    user_id=user_id,
    agent_id="assistant_agent",
    limit=5,
    add_profile=True  # Include profile
)

print(f"✓ Search with profile completed")
print(f"  - Results count: {len(results.get('results', []))}")
print(f"  - Profile included: {'profile_content' in results}")

if 'profile_content' in results:
    print(f"\n  User Profile:")
    print(f"  {results['profile_content']}")

print(f"\n  Search Results:")
for i, result in enumerate(results.get('results', []), 1):
    print(f"  {i}. {result.get('memory', '')} (score: {result.get('score', 0):.2f})")

Run this code:

python user_profile_example.py

Expected output:

✓ Search with profile completed
  - Results count: 1
  - Profile included: True

  User Profile:
  Name: Diana. Profession: UX designer. Interests: Creating beautiful interfaces.

  Search Results:
  1. Works as a UX designer, loves creating beautiful interfaces (score: 0.92)

Step 7: Extract Profile in Native Language

You can specify a native language for profile extraction, ensuring the profile is written in the user's preferred language regardless of the conversation language:

# user_profile_example.py
from powermem import UserMemory, auto_config

config = auto_config()
user_memory = UserMemory(config=config)

# Example 1: English conversation, Chinese profile
conversation_en = [
    {"role": "user", "content": "I am a software engineer working in Beijing. I love drinking tea and reading books."},
    {"role": "assistant", "content": "That sounds great!"}
]

result_zh = user_memory.add(
    messages=conversation_en,
    user_id="user_bilingual_001",
    native_language="zh"  # Extract profile in Chinese
)

print("✓ English conversation processed")
if result_zh.get('profile_content'):
    print(f"  - Profile (Chinese): {result_zh['profile_content']}")

# Example 2: Chinese conversation, English profile
conversation_zh = [
    {"role": "user", "content": "I'm 25 years old, working at Microsoft in Seattle."},
    {"role": "assistant", "content": "Nice to meet you"}
]

result_en = user_memory.add(
    messages=conversation_zh,
    user_id="user_bilingual_002",
    native_language="en"  # Extract profile in English
)

print("\n✓ Chinese conversation processed")
if result_en.get('profile_content'):
    print(f"  - Profile (English): {result_en['profile_content']}")

# Example 3: Structured topics with native language
result_topics = user_memory.add(
    messages="I'm 25 years old, working at Microsoft in Seattle.",
    user_id="user_bilingual_003",
    profile_type="topics",
    native_language="zh"  # Topic values in Chinese, keys remain English
)

print("\n✓ Structured topics extracted")
if result_topics.get('topics'):
    print(f"  - Topics: {result_topics['topics']}")

Run this code:

python user_profile_example.py

Supported Language Codes:

Code	Language	Code	Language
zh	Chinese	en	English
ja	Japanese	ko	Korean
fr	French	de	German
es	Spanish	it	Italian
pt	Portuguese	ru	Russian
ar	Arabic	hi	Hindi
th	Thai	vi	Vietnamese

Step 8: Delete User Profile

Delete a user profile:

# user_profile_example.py
from powermem import UserMemory, auto_config

config = auto_config()
user_memory = UserMemory(config=config, agent_id="assistant_agent")

user_id = "user_007"

# Add conversation and create profile
user_memory.add(
    messages=[
        {"role": "user", "content": "I'm Grace, a teacher."},
        {"role": "assistant", "content": "Nice to meet you!"}
    ],
    user_id=user_id
)

# Get profile to confirm it exists
profile = user_memory.profile(user_id=user_id)
if profile:
    profile_id = profile.get('id')
    print(f"✓ Profile exists with ID: {profile_id}")
    
    # Delete the profile
    deleted = user_memory.delete_profile(
        user_id=user_id
    )
    
    if deleted:
        print(f"✓ Profile deleted successfully")
    else:
        print(f"✗ Failed to delete profile")
    
    # Verify deletion
    profile_after = user_memory.profile(user_id=user_id)
    if not profile_after:
        print("✓ Profile confirmed deleted")
    else:
        print("✗ Profile still exists")

Run this code:

python user_profile_example.py

Expected output:

✓ Profile exists with ID: 1234567890123456789
✓ Profile deleted successfully
✓ Profile confirmed deleted

Complete Example

Here's a complete example combining all the features:

# complete_user_profile_example.py
from powermem import UserMemory, auto_config

def main():
    # Load configuration
    config = auto_config()
    
    # Initialize UserMemory
    user_memory = UserMemory(config=config, agent_id="demo_agent")
    user_id = "demo_user"
    
    print("=" * 60)
    print("UserMemory Profile Management Example")
    print("=" * 60)
    
    # Step 1: Add initial conversation
    print("\n1. Adding initial conversation...")
    conversation1 = [
        {"role": "user", "content": "Hi, I'm Alex, a 32-year-old data scientist from New York. I specialize in machine learning and love reading tech blogs."},
        {"role": "assistant", "content": "Nice to meet you, Alex! That's fascinating."}
    ]
    
    result1 = user_memory.add(
        messages=conversation1,
        user_id=user_id,
        run_id="session_001"
    )
    
    print(f"   ✓ Profile extracted: {result1.get('profile_extracted', False)}")
    if result1.get('profile_content'):
        print(f"   Profile: {result1['profile_content']}")
    
    # Step 2: Update profile with more information
    print("\n2. Updating profile with more conversations...")
    conversation2 = [
        {"role": "user", "content": "I also enjoy hiking and photography on weekends."},
        {"role": "assistant", "content": "Those are great hobbies!"}
    ]
    
    result2 = user_memory.add(
        messages=conversation2,
        user_id=user_id,
        run_id="session_002"
    )
    
    print(f"   ✓ Profile updated: {result2.get('profile_extracted', False)}")
    
    # Step 3: Get full profile
    print("\n3. Retrieving full profile...")
    profile = user_memory.profile(
        user_id=user_id
    )
    
    if profile:
        print(f"   ✓ Profile ID: {profile.get('id')}")
        print(f"   Profile content: {profile.get('profile_content', '')}")
        print(f"   Last updated: {profile.get('updated_at')}")
    
    # Step 4: Search with profile
    print("\n4. Searching memories with profile...")
    search_results = user_memory.search(
        query="user interests and hobbies",
        user_id=user_id,
        limit=5,
        add_profile=True
    )
    
    if 'profile_content' in search_results:
        print(f"   User Profile: {search_results['profile_content']}")
    
    print(f"\n   Found {len(search_results.get('results', []))} memories:")
    for i, result in enumerate(search_results.get('results', []), 1):
        print(f"   {i}. {result.get('memory', '')} (score: {result.get('score', 0):.2f})")
    
    # Step 5: Cleanup (optional)
    print("\n5. Cleaning up...")
    # Note: In production, you might want to keep the profiles
    # deleted = user_memory.delete_profile(user_id=user_id)
    # if deleted:
    #     print("   ✓ Profile deleted")
    
    print("\n" + "=" * 60)
    print("Example completed successfully!")
    print("=" * 60)

if __name__ == "__main__":
    main()

Run this code:

python complete_user_profile_example.py

Best Practices

Always provide user_id: The add() method requires user_id; ensure each user has a unique identifier
Use agent_id to distinguish agents: In multi-agent scenarios, use agent_id to separate profiles and memories across agents
Use run_id appropriately: Use run_id to distinguish sessions or runs for more precise profile management
Check profiles regularly: Use profile() to periodically check and ensure profile information remains accurate
Include profile when needed: When additional context is useful, set add_profile=True to include the user profile in search results
Handle empty profiles: If profile() returns {}, no profile has been extracted yet; call add() with conversation data first
Profile updates: Profiles are automatically updated when you add new conversations with the same user_id, agent_id, and run_id combination
Message filtering: By default, only user role messages are used for profile extraction (assistant messages are excluded). You can customize this with include_roles and exclude_roles parameters
Query rewrite (optional): Enable query_rewrite.enabled=True to let search() rewrite queries using profile content; it falls back to the original query when unavailable

Step 8: Filter Messages by Roles

By default, UserMemory.add() only extracts profile information from user messages. You can customize which message roles are used for profile extraction:

# user_profile_example.py
from powermem import UserMemory, auto_config

config = auto_config()
user_memory = UserMemory(config=config)

# Conversation with multiple roles
conversation = [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "Hi, I'm Frank, a doctor from Boston."},
    {"role": "assistant", "content": "Nice to meet you, Frank!"},
    {"role": "tool", "content": "Weather data: Boston, 72°F"}
]

# Default behavior: only include 'user' messages, exclude 'assistant' messages
result = user_memory.add(
    messages=conversation,
    user_id="user_008"
)
print("Default filtering (user only):")
print(f"  Profile: {result.get('profile_content', 'N/A')}")

# Include all messages (disable filtering)
result = user_memory.add(
    messages=conversation,
    user_id="user_009",
    include_roles=None,  # or []
    exclude_roles=None   # or []
)
print("\nNo filtering (all roles):")
print(f"  Profile: {result.get('profile_content', 'N/A')}")

# Custom filtering: include user and system, exclude tool
result = user_memory.add(
    messages=conversation,
    user_id="user_010",
    include_roles=["user", "system"],
    exclude_roles=["tool"]
)
print("\nCustom filtering (user + system, exclude tool):")
print(f"  Profile: {result.get('profile_content', 'N/A')}")

Run this code:

python user_profile_example.py

Expected output:

Default filtering (user only):
  Profile: Name: Frank. Profession: Doctor. Location: Boston.

No filtering (all roles):
  Profile: Name: Frank. Profession: Doctor. Location: Boston.

Custom filtering (user + system, exclude tool):
  Profile: Name: Frank. Profession: Doctor. Location: Boston.

UserMemory Guide - Detailed guide on UserMemory features
Getting Started - Learn the basics of PowerMem
Multi-Agent Guide - Using multiple agents with Memory

Prerequisites​

Understanding User Profile Management​

Step 1: Initialize UserMemory​

Step 2: Add Conversation and Extract Profile​

Step 3: Update Profile with More Conversations​

Step 4: Get User Profile​

Step 5: Search Memories Without Profile​

Optional: Query Rewrite with User Profile​

Step 6: Search Memories With Profile​

Step 7: Extract Profile in Native Language​

Step 8: Delete User Profile​

Complete Example​

Best Practices​

Step 8: Filter Messages by Roles​

Related Documents​

Prerequisites

Understanding User Profile Management

Step 1: Initialize UserMemory

Step 2: Add Conversation and Extract Profile

Step 3: Update Profile with More Conversations

Step 4: Get User Profile

Step 5: Search Memories Without Profile

Optional: Query Rewrite with User Profile

Step 6: Search Memories With Profile

Step 7: Extract Profile in Native Language

Step 8: Delete User Profile

Complete Example

Best Practices

Step 8: Filter Messages by Roles

Related Documents