LangGraph + MemorySync

Overview

The langgraph surface of the LangChain packages ships a genuine BaseStore implementation compiled straight into any graph via store= (its values live in MemorySync framework state, apart from the user’s memories), pre-model context injection through create_react_agent hooks in both languages and create_agent middleware in Python, a persist node that sends the latest user message for fact extraction (a re-run node or retried model call is recognised, so nothing is extracted twice), and a graph-callable search tool — in Python and Node.js, sharing one wire format.

Setup status and requirements

Supported version

langchain-memorysync[langgraph] 1.2.0 (PyPI) / memorysync-langchain 1.2.0 (npm)

Last setup review

2026-10-01

Permissions

A MemorySync API key with read and write scopes; store deletes need delete permission.

Limits

Requires langgraph 1.2+ (Python 3.10+) or @langchain/langgraph 0.2+ (Node 18+). Thread state belongs in a checkpointer such as langgraph-checkpoint-postgres; this integration is the long-term memory half.

Capabilities

• LangGraph BaseStore with cross-thread durable state — one live value per key
• Semantic store search with zero embedding configuration
• Pre-model context injection for create_react_agent, plus create_agent middleware in Python
• Persist node: the latest user message goes to fact extraction, never twice
• Graph-callable memory search tool that never raises
• One wire format across Python and Node.js — data written by one is read by the other

Quick Start

from langchain.agents import create_agent
from langchain_memorysync.langgraph import (
    MemorySyncMemoryMiddleware,
    MemorySyncStore,
)

agent = create_agent(
    model,
    tools=tools,
    middleware=[MemorySyncMemoryMiddleware(user_id="customer-7")],
    store=MemorySyncStore(),
)

Use Cases

Explore More