memorylayer-sdk-langchain-python

MemoryLayer LangChain Integration

LangChain memory integration for MemoryLayer.ai - Persistent memory for AI agents.

Installation

pip install memorylayer-langchain

Overview

This package provides LangChain-compatible memory classes that use MemoryLayer as the backend, giving you:

• Persistent Memory - Memory survives across agent runs and application restarts
• Stable API - Consistent interface regardless of LangChain version changes
• Rich Memory Types - Semantic, episodic, procedural, and working memory support
• Drop-in Replacement - Works with LCEL and legacy chains
• Session Isolation - Multiple conversations tracked independently

Quick Start

LCEL Chains (Recommended)

from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.runnables.history import RunnableWithMessageHistory
from langchain_openai import ChatOpenAI

from memorylayer_langchain import MemoryLayerChatMessageHistory

# Create a chat chain with message history
prompt = ChatPromptTemplate.from_messages([
    ("system", "You are a helpful assistant."),
    MessagesPlaceholder(variable_name="history"),
    ("human", "{input}"),
])

llm = ChatOpenAI(model="gpt-4")
chain = prompt | llm

# Wrap with persistent history
chain_with_history = RunnableWithMessageHistory(
    runnable=chain,
    get_session_history=lambda session_id: MemoryLayerChatMessageHistory(
        session_id=session_id,
        base_url="http://localhost:61001",
        api_key="your-api-key",
        workspace_id="ws_123",
    ),
    input_messages_key="input",
    history_messages_key="history",
)

# Use with any session - history persists automatically
response = chain_with_history.invoke(
    {"input": "Hello! My name is Alice."},
    config={"configurable": {"session_id": "user_alice_session_1"}},
)

# Later, even after restart, Alice's context is remembered
response = chain_with_history.invoke(
    {"input": "What's my name?"},
    config={"configurable": {"session_id": "user_alice_session_1"}},
)

Legacy Chains

from langchain.chains import ConversationChain
from langchain_openai import ChatOpenAI

from memorylayer_langchain import MemoryLayerMemory

# Create persistent memory
memory = MemoryLayerMemory(
    session_id="customer_support_session_1",
    base_url="http://localhost:61001",
    api_key="your-api-key",
    workspace_id="ws_123",
)

# Use with ConversationChain
llm = ChatOpenAI(model="gpt-4")
chain = ConversationChain(llm=llm, memory=memory)

# Conversation persists across chain invocations and restarts
chain.run("Hi, I need help with my order #12345")
chain.run("It hasn't arrived yet")

Features

MemoryLayerChatMessageHistory

Drop-in replacement for LangChain chat history, designed for LCEL chains.

from memorylayer_langchain import MemoryLayerChatMessageHistory

history = MemoryLayerChatMessageHistory(
    session_id="conversation_1",
    base_url="http://localhost:61001",
    api_key="your-api-key",
    workspace_id="ws_123",
)

# Add messages
history.add_user_message("Hello!")
history.add_ai_message("Hi there! How can I help?")

# Retrieve all messages
messages = history.messages
for msg in messages:
    print(f"{msg.type}: {msg.content}")

# Clear history
history.clear()

MemoryLayerMemory

LangChain BaseMemory implementation for legacy chains. Drop-in replacement for ConversationBufferMemory.

from memorylayer_langchain import MemoryLayerMemory

memory = MemoryLayerMemory(
    session_id="user_123_conversation",
    base_url="http://localhost:61001",
    api_key="your-api-key",
    workspace_id="ws_123",
    return_messages=False,  # True for message objects
    human_prefix="User",
    ai_prefix="Assistant",
)

# Save conversation turn
memory.save_context(
    inputs={"input": "What's Python?"},
    outputs={"output": "Python is a programming language."},
)

# Load memory variables
history = memory.load_memory_variables({})
print(history["history"])
# Output: User: What's Python?
#         Assistant: Python is a programming language.

MemoryLayerConversationSummaryMemory

Returns AI-generated summaries instead of full conversation history. Useful for long conversations that would exceed context windows.

from memorylayer_langchain import MemoryLayerConversationSummaryMemory

memory = MemoryLayerConversationSummaryMemory(
    session_id="long_conversation",
    base_url="http://localhost:61001",
    api_key="your-api-key",
    workspace_id="ws_123",
    max_tokens=500,
    summary_prompt="Summarize the key points from this conversation.",
)

# After many conversation turns...
summary = memory.load_memory_variables({})
print(summary["history"])  # Concise AI-generated summary

Configuration Options

Common Parameters

Parameter	Description	Default
session_id	Unique identifier for the conversation session	Required
base_url	MemoryLayer API base URL	http://localhost:61001
api_key	API key for authentication	None
workspace_id	Workspace ID for multi-tenant isolation	None
timeout	Request timeout in seconds	30.0
memory_tags	Additional tags for stored memories	[]

MemoryLayerMemory Options

Parameter	Description	Default
memory_key	Key for memory variables	"history"
return_messages	Return message objects vs string	False
human_prefix	Prefix for human messages	"Human"
ai_prefix	Prefix for AI messages	"AI"
input_key	Custom input key	None
output_key	Custom output key	None

MemoryLayerConversationSummaryMemory Options

Parameter	Description	Default
max_tokens	Maximum tokens in summary	500
summary_prompt	Custom summarization prompt	Built-in
include_sources	Include source memory IDs	False

Advanced Usage

Custom Memory Tags

Tag messages for cross-session filtering and organization:

history = MemoryLayerChatMessageHistory(
    session_id="support_ticket_456",
    base_url="http://localhost:61001",
    api_key="your-api-key",
    workspace_id="ws_123",
    memory_tags=["customer:enterprise", "topic:billing", "priority:high"],
)

Multi-Session Management

Track multiple conversations independently within the same workspace:

# Session for user 1
user1_history = MemoryLayerChatMessageHistory(
    session_id="user_1_main",
    base_url="http://localhost:61001",
    api_key="your-api-key",
    workspace_id="ws_123",
)

# Session for user 2 - completely isolated
user2_history = MemoryLayerChatMessageHistory(
    session_id="user_2_main",
    base_url="http://localhost:61001",
    api_key="your-api-key",
    workspace_id="ws_123",
)

Streaming Support

LCEL chains with RunnableWithMessageHistory support streaming natively:

chain_with_history = RunnableWithMessageHistory(
    runnable=chain,
    get_session_history=lambda session_id: MemoryLayerChatMessageHistory(
        session_id=session_id,
        base_url="http://localhost:61001",
        api_key="your-api-key",
        workspace_id="ws_123",
    ),
    input_messages_key="input",
    history_messages_key="history",
)

# Stream the response
for chunk in chain_with_history.stream(
    {"input": "Tell me a story"},
    config={"configurable": {"session_id": "story_session"}},
):
    print(chunk.content, end="", flush=True)

Custom Input/Output Keys

Match your chain's key names:

memory = MemoryLayerMemory(
    session_id="qa_session",
    base_url="http://localhost:61001",
    api_key="your-api-key",
    workspace_id="ws_123",
    input_key="question",
    output_key="answer",
    memory_key="chat_history",
)

memory.save_context(
    inputs={"question": "What is Python?"},
    outputs={"answer": "Python is a programming language."},
)

result = memory.load_memory_variables({})
print(result["chat_history"])

Synchronous Client

For direct API access without LangChain abstractions:

from memorylayer_langchain import SyncMemoryLayerClient, sync_client

# Using context manager
with sync_client(
    base_url="http://localhost:61001",
    api_key="your-api-key",
    workspace_id="ws_123",
) as client:
    # Store a memory
    memory = client.remember(
        content="User prefers Python for backend development",
        type="semantic",
        importance=0.8,
        tags=["preferences", "programming"],
    )

    # Search memories
    results = client.recall(
        query="what programming language does the user prefer?",
        limit=5,
    )

    # Get a summary
    reflection = client.reflect(
        query="summarize user's technology preferences",
        max_tokens=300,
    )

Migration from LangChain Memory

From ConversationBufferMemory

# Before (LangChain built-in - not persistent)
from langchain.memory import ConversationBufferMemory
memory = ConversationBufferMemory()

# After (MemoryLayer - persistent)
from memorylayer_langchain import MemoryLayerMemory
memory = MemoryLayerMemory(
    session_id="my_session",
    base_url="http://localhost:61001",
    api_key="your-api-key",
    workspace_id="ws_123",
)

From ConversationSummaryMemory

# Before (LangChain built-in - not persistent)
from langchain.memory import ConversationSummaryMemory
memory = ConversationSummaryMemory(llm=llm)

# After (MemoryLayer - persistent)
from memorylayer_langchain import MemoryLayerConversationSummaryMemory
memory = MemoryLayerConversationSummaryMemory(
    session_id="my_session",
    base_url="http://localhost:61001",
    api_key="your-api-key",
    workspace_id="ws_123",
)

Why MemoryLayer?

Problem: LangChain Memory Doesn't Persist

Standard LangChain memory is lost when your application restarts:

# LangChain's built-in memory
memory = ConversationBufferMemory()
chain = ConversationChain(llm=llm, memory=memory)
chain.run("My name is Alice")  # Memory stored in RAM

# Application restarts... memory is gone!

Solution: MemoryLayer Provides True Persistence

# MemoryLayer integration
memory = MemoryLayerMemory(session_id="alice_session", ...)
chain = ConversationChain(llm=llm, memory=memory)
chain.run("My name is Alice")  # Memory stored in MemoryLayer

# Application restarts... memory is preserved!
memory2 = MemoryLayerMemory(session_id="alice_session", ...)
# Alice's conversation history is still available

Additional Benefits

• Stable API - LangChain memory interfaces change frequently. MemoryLayer provides a stable abstraction.
• Cross-Platform - Access the same memories from Python, TypeScript, or any HTTP client.
• Rich Memory Types - Beyond simple chat history: semantic, episodic, procedural memories with relationships.
• Built-in Search - Semantic search across all stored memories.
• Reflection - AI-powered synthesis and summarization of memories.

Development

# Install development dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Type checking
mypy src/memorylayer_langchain

# Linting
ruff check src/memorylayer_langchain

Examples

See the examples/ directory for complete working examples:

• lcel_example.py - Modern LCEL chains with RunnableWithMessageHistory
• legacy_chain_example.py - Legacy ConversationChain integration
• summary_memory_example.py - Conversation summary memory usage

License

Apache 2.0 License -- see LICENSE for details.

Links

• MemoryLayer Documentation
• LangChain Documentation
• GitHub
• Homepage