Skip to main content

​
Memory Integration

Chat automatically manages conversation memory through the Memory system. It handles conversation history, summarization, and user profiles.

​
Memory Features

Chat integrates with Memory to provide:
  • Full Session Memory: Complete conversation history stored in InteractionSession.chat_history
  • Summary Memory: Automatic conversation summarization (requires model configuration)
  • User Analysis Memory: User profile tracking and analysis (requires model configuration)

​
Basic Memory Usage

from upsonic import Agent, Chat

agent = Agent("openai/gpt-4o")

chat = Chat(
    session_id="session1",
    user_id="user1",
    agent=agent,
    full_session_memory=True
)

# Chat automatically uses memory
await chat.invoke("My name is Alice")
await chat.invoke("What's my name?")  # Agent remembers from previous message

​
Advanced Memory Configuration

from upsonic import Agent, Chat

agent = Agent("openai/gpt-4o")

chat = Chat(
    session_id="session1",
    user_id="user1",
    agent=agent,
    full_session_memory=True,
    summary_memory=True,  # Requires model (uses agent.model)
    user_analysis_memory=True,  # Requires model (uses agent.model)
    num_last_messages=50,  # Limits to last 50 message pairs
    feed_tool_call_results=True  # Include tool calls in history
)
Note: summary_memory and user_analysis_memory require the agent’s model to be configured. They use agent.model for LLM-based summarization and user trait extraction.

​
Memory with Custom Storage

from upsonic import Agent, Chat
from upsonic.storage.providers import SqliteStorage

storage = SqliteStorage("sessions", "profiles", "chat.db")
agent = Agent("openai/gpt-4o")

chat = Chat(
    session_id="session1",
    user_id="user1",
    agent=agent,
    storage=storage,
    full_session_memory=True,
    summary_memory=True
)

# Memory persists across sessions
await chat.invoke("Remember: I prefer dark mode")
# Later, in a new session with same user_id
chat2 = Chat(
    session_id="session2",
    user_id="user1",
    agent=agent,
    storage=storage,
    full_session_memory=True,
    summary_memory=True
)
await chat2.invoke("What's my preference?")  # Agent remembers

​
User Profile Schema

Use a custom Pydantic model for structured user profiles: 
from pydantic import BaseModel
from upsonic import Agent, Chat

class UserProfile(BaseModel):
    name: str
    preferences: dict

agent = Agent("openai/gpt-4o")

chat = Chat(
    session_id="session1",
    user_id="user1",
    agent=agent,
    user_profile_schema=UserProfile,
    user_analysis_memory=True
)

​
Dynamic Profile Schema

Enable automatic schema generation from conversations: 
from upsonic import Agent, Chat

agent = Agent("openai/gpt-4o")

chat = Chat(
    session_id="session1",
    user_id="user1",
    agent=agent,
    dynamic_user_profile=True,  # Automatically generate schema fields
    user_analysis_memory=True  # Required for dynamic profiles
)

# The schema will be automatically generated based on user conversations
# Fields are extracted as the user mentions information
When dynamic_user_profile=True, the system automatically identifies 2-5 relevant traits from conversations and creates a dynamic schema. This is useful when you don’t know the user profile structure in advance. Note: If both dynamic_user_profile=True and user_profile_schema are provided, the dynamic schema takes precedence.

​
Memory Modes

  • update: Incrementally update user profiles, merging new traits with existing ones (default)
  • replace: Replace user profiles completely on each update
from upsonic import Agent, Chat

agent = Agent("openai/gpt-4o")

chat = Chat(
    session_id="session1",
    user_id="user1",
    agent=agent,
    user_analysis_memory=True,
    user_memory_mode="update"  # or "replace"
)

​
Limiting Message History

The num_last_messages parameter limits conversation history to the last N message pairs (request-response pairs): 
from upsonic import Agent, Chat

agent = Agent("openai/gpt-4o")

chat = Chat(
    session_id="session1",
    user_id="user1",
    agent=agent,
    full_session_memory=True,
    num_last_messages=20  # Keep last 20 request-response pairs
)
This helps manage context size and reduce token costs for long conversations.

​
Tool Call Filtering

By default, tool calls are excluded from memory. Enable them with feed_tool_call_results: 
from upsonic import Agent, Chat

agent = Agent("openai/gpt-4o")

chat = Chat(
    session_id="session1",
    user_id="user1",
    agent=agent,
    full_session_memory=True,
    feed_tool_call_results=True  # Include tool execution results
)
When False (default), tool-related messages are filtered from chat_history to keep memory focused on user interactions.