Overview
Attribution in Memori allows you to associate memories with specific entities (users, customers, agents) and processes (applications, workflows). This enables you to:
- Isolate memories per user or customer
- Track memories across different applications
- Organize memories by context or workflow
attribution()
Set entity and process identifiers for memory attribution.
Parameters
entity_id
str | int | None
default:"None"
Unique identifier for the entity. Can be a user ID, customer ID, agent ID, or any unique identifier.Constraints:
- Automatically converted to string
- Maximum length: 100 characters
- Raises
RuntimeError if length exceeds 100 characters
Examples:
"user-123""customer-abc-456""agent-support-001"12345 (converted to "12345")
process_id
str | int | None
default:"None"
Identifier for the process or application context. Use this to distinguish between different applications or workflows.Constraints:
- Automatically converted to string
- Maximum length: 100 characters
- Raises
RuntimeError if length exceeds 100 characters
Examples:
"my-app""customer-support""sales-agent""onboarding-flow"
Returns
Returns the Memori instance for method chaining
Usage Examples
Basic Attribution
from memori import Memori
mem = Memori()
mem.attribution(entity_id="user-123", process_id="my-app")
Method Chaining
mem = Memori().attribution(
entity_id="customer-456",
process_id="support-agent"
).new_session()
Customer Support Context
from memori import Memori
from sqlalchemy import create_engine
from sqlalchemy.orm import sessionmaker
engine = create_engine("postgresql://localhost/memori")
Session = sessionmaker(bind=engine)
mem = Memori(conn=Session)
# Set attribution for a specific customer and support process
mem.attribution(
entity_id="customer-789",
process_id="technical-support"
)
# All memories will be associated with this customer
facts = mem.recall("What issues has this customer reported?")
Multi-Tenant Application
def get_memori_for_user(user_id: str, tenant_id: str):
mem = Memori(conn=get_db_session())
mem.attribution(
entity_id=f"{tenant_id}:{user_id}",
process_id="saas-platform"
)
return mem
# Usage
user_mem = get_memori_for_user("user-123", "tenant-acme")
facts = user_mem.recall("user preferences")
Numeric IDs
# Numeric IDs are automatically converted to strings
mem.attribution(
entity_id=12345, # Converted to "12345"
process_id=67890 # Converted to "67890"
)
Updating Attribution
mem = Memori()
# Initial attribution
mem.attribution(entity_id="user-123", process_id="app-1")
# Update attribution for a different context
mem.attribution(entity_id="user-456", process_id="app-2")
# Update only entity_id
mem.attribution(entity_id="user-789")
Attribution and Recall
When you set attribution, the entity_id is used to filter memories during recall:
mem = Memori(conn=Session)
mem.attribution(entity_id="user-123", process_id="my-app")
# This recall only returns memories associated with "user-123"
facts = mem.recall("What are my preferences?")
Attribution and Cloud Storage
When using Memori’s cloud storage, attribution is sent with API requests:
import os
os.environ["MEMORI_API_KEY"] = "your-api-key"
mem = Memori() # Uses cloud storage
mem.attribution(entity_id="user-123", process_id="my-app")
# Attribution is included in cloud recall requests
facts = mem.recall("user preferences")
Best Practices
1. Set Attribution Early
Always set attribution before registering LLMs or performing recall:
mem = Memori(conn=Session)
mem.attribution(entity_id="user-123", process_id="my-app")
mem.llm.register(openai_client)
2. Use Consistent Entity IDs
Use the same entity_id format across your application:
# Good: Consistent format
mem.attribution(entity_id="user-123")
mem.attribution(entity_id="user-456")
# Avoid: Inconsistent formats
mem.attribution(entity_id="user-123")
mem.attribution(entity_id="customer_456") # Different format
3. Use Process ID for Context
Use process_id to distinguish between different application contexts:
# Support agent
support_mem = Memori().attribution(
entity_id="customer-123",
process_id="support-chat"
)
# Sales agent
sales_mem = Memori().attribution(
entity_id="customer-123",
process_id="sales-inquiry"
)
4. Validate ID Length
Ensure IDs don’t exceed 100 characters:
def safe_attribution(mem: Memori, entity_id: str, process_id: str):
if len(entity_id) > 100:
raise ValueError("entity_id too long")
if len(process_id) > 100:
raise ValueError("process_id too long")
mem.attribution(entity_id=entity_id, process_id=process_id)
Error Handling
try:
mem.attribution(
entity_id="a" * 101, # Too long!
process_id="my-app"
)
except RuntimeError as e:
print(f"Error: {e}") # "entity_id cannot be greater than 100 characters"
Internal Behavior
The attribution() method stores IDs in the config:
mem = Memori()
mem.attribution(entity_id="user-123", process_id="my-app")
# Accessible via config
print(mem.config.entity_id) # "user-123"
print(mem.config.process_id) # "my-app"
See Also