langswarm-memory 0.1.0

Enterprise-grade conversational memory for AI agents

AgentMem

Enterprise-grade conversational memory for AI agents

AgentMem provides session-based conversation management, multiple storage backends, and seamless integration with major LLM providers.

---

Features

✨ Session Management - Organize conversations with persistent sessions
💾 Multiple Backends - SQLite, Redis, In-Memory, and vector stores
🔄 Auto-Summarization - Automatic conversation compression when limits reached
📊 Token Management - Keep context within model limits
🎯 LLM Provider Integration - Native OpenAI and Anthropic format support
⚡ Async First - Built for high-performance async applications

---

Quick Start

Installation

pip install agentmem

Optional dependencies:

pip install agentmem[redis]      # Redis backend
pip install agentmem[vector]     # Vector store support
pip install agentmem[chromadb]   # ChromaDB vector store
pip install agentmem[all]        # All optional dependencies

Basic Usage

import asyncio
from agentmem import create_memory_manager, Message, MessageRole

async def main():
    # Create memory manager with SQLite backend
    manager = create_memory_manager("sqlite", db_path="conversations.db")
    await manager.backend.connect()
    
    # Create a session
    session = await manager.create_session(user_id="user123")
    
    # Add messages
    await session.add_message(Message(
        role=MessageRole.USER,
        content="Hello! What's the capital of France?"
    ))
    
    await session.add_message(Message(
        role=MessageRole.ASSISTANT,
        content="The capital of France is Paris."
    ))
    
    # Get conversation history
    messages = await session.get_messages()
    print(f"Conversation has {len(messages)} messages")
    
    # Get recent context (token-limited)
    context = await session.get_recent_context(max_tokens=2000)

asyncio.run(main())

With OpenAI

import asyncio
from openai import AsyncOpenAI
from agentmem import create_memory_manager, Message, MessageRole

async def chat_with_memory(user_message: str, user_id: str = "default"):
    # Initialize
    openai_client = AsyncOpenAI()
    manager = create_memory_manager("sqlite", db_path="chat.db")
    await manager.backend.connect()
    
    # Get or create session
    session = await manager.get_or_create_session(
        session_id=f"session_{user_id}",
        user_id=user_id
    )
    
    # Add user message
    await session.add_message(Message(
        role=MessageRole.USER,
        content=user_message
    ))
    
    # Get conversation history
    messages = await session.get_messages()
    
    # Convert to OpenAI format
    openai_messages = [msg.to_openai_format() for msg in messages]
    
    # Get AI response
    response = await openai_client.chat.completions.create(
        model="gpt-3.5-turbo",
        messages=openai_messages
    )
    
    assistant_message = response.choices[0].message.content
    
    # Save assistant response
    await session.add_message(Message(
        role=MessageRole.ASSISTANT,
        content=assistant_message
    ))
    
    return assistant_message

# Run
asyncio.run(chat_with_memory("What's the weather like?"))

---

Backends

In-Memory (Development)

from agentmem import InMemoryBackend

backend = InMemoryBackend()
await backend.connect()

SQLite (Production)

from agentmem import SQLiteBackend

backend = SQLiteBackend(db_path="memory.db")
await backend.connect()

Redis (Distributed)

from agentmem import RedisBackend

backend = RedisBackend(
    host="localhost",
    port=6379,
    password="secret"  # optional
)
await backend.connect()

---

Advanced Features

Auto-Summarization

Automatically compress conversations when they exceed thresholds:

session = await manager.create_session(
    user_id="user123",
    max_messages=100,
    auto_summarize=True,
    summary_threshold=50  # Summarize after 50 messages
)

Token Management

Keep context within model limits:

# Get recent messages that fit within token budget
context = await session.get_recent_context(max_tokens=2000)

# Convert to your LLM format
openai_messages = [msg.to_openai_format() for msg in context]
anthropic_messages = [msg.to_anthropic_format() for msg in context]

Session Metadata

Track additional information:

session = await manager.create_session(
    user_id="user123",
    agent_id="support_agent",
    workflow_id="customer_support",
    tags=["support", "billing"],
    properties={"priority": "high", "department": "sales"}
)

Clean Up Expired Sessions

# Cleanup sessions that have expired
deleted_count = await manager.cleanup_expired_sessions()
print(f"Cleaned up {deleted_count} expired sessions")

---

Configuration Patterns

Development

# Fast, non-persistent
manager = create_memory_manager("memory")

Testing

# In-memory SQLite
manager = create_memory_manager("sqlite", db_path=":memory:")

Production

# Persistent SQLite with custom settings
manager = create_memory_manager("sqlite", 
    db_path="/var/app/conversations.db",
    pool_size=10,
    timeout=30
)

Distributed

# Redis for multi-instance deployments
manager = create_memory_manager("redis",
    host="redis.example.com",
    port=6379,
    db=0,
    password=os.getenv("REDIS_PASSWORD")
)

---

API Reference

Core Classes

• Message: Universal message format with role, content, and metadata
• SessionMetadata: Session configuration and state
• ConversationSummary: Automatically generated conversation summaries
• MemoryUsage: Memory usage statistics

Interfaces

• IMemorySession: Session management interface
• IMemoryBackend: Backend storage interface
• IMemoryManager: High-level memory management interface

Factory Functions

• create_memory_manager(): Create a memory manager with specified backend
• create_memory_backend(): Create a backend instance directly

---

Integration Examples

LangChain

from langchain.memory import ConversationBufferMemory
from agentmem import create_memory_manager

# Use agentmem as LangChain memory backend
manager = create_memory_manager("sqlite")
# ... integrate with LangChain chains

LlamaIndex

from llama_index import ChatMemoryBuffer
from agentmem import create_memory_manager

# Use agentmem with LlamaIndex
manager = create_memory_manager("sqlite")
# ... integrate with LlamaIndex agents

---

Roadmap

Phase 1 (Current) - Conversational Memory

• ✅ Session management
• ✅ Multiple backends (SQLite, Redis, In-Memory)
• ✅ Auto-summarization
• ✅ Token management
• ✅ LLM provider integration

Phase 2 (Planned) - Agent Memory

• 🔜 6 memory types (Working, Episodic, Semantic, Procedural, Emotional, Preference)
• 🔜 Personalization engine
• 🔜 Context compression strategies
• 🔜 Memory analytics and optimization
• 🔜 Long-term semantic memory with vector search

---

Contributing

Contributions are welcome! Please see CONTRIBUTING.md for details.

1. Fork the repository
2. Create your feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

---

License

Apache License 2.0 - see LICENSE file for details.

---

Support

• Documentation: https://github.com/aekdahl/agentmem
• Issues: https://github.com/aekdahl/agentmem/issues
• Discussions: https://github.com/aekdahl/agentmem/discussions

---

Acknowledgments

AgentMem is extracted from LangSwarm, a multi-agent AI orchestration framework. It represents Phase 1 of the memory system, focusing on conversational memory management.

---

Built with ❤️ for the AI agent community