LangChain Cross-Session Storage — Store
Checkpointer solves the problem of "memory within a single conversation." But if you need to share data across different conversations—such as user preferences or learning progress—you need to useStore。
Checkpointer vs Store
| Dimension | Checkpointer | Store |
|---|---|---|
| Scope | Single conversation thread (thread_id) | Across all conversation threads |
| Data type | Agent state snapshot (automatically managed) | Arbitrary key-value data (manually managed) |
| Typical use cases | Multi-turn conversation memory | User preferences, knowledge base, configuration |
| Data organization | thread_id → checkpoint | (namespace, key) → value |
Basic Store Operations
Store UsageNamespace + keyto organize data in a hierarchical structure:
Example
from langgraph.store.memory import InMemoryStore
store = InMemoryStore()
# Write data: put(namespace, key, value)
# namespace is a tuple, key is a string, value is a dictionary
store.put(
("users", "user_001"), # Namespace
"preferences", # Key
{ # Value
"theme": "dark",
"language": "zh-CN",
"level": "Beginner",
}
)
store.put(
("users", "user_001"),
"progress",
{
"completed_courses": ["HTML Basics", "Python Basics"],
"total_hours": 35,
}
)
# Read data: get(namespace, key)
prefs = store.get(("users", "user_001"), "preferences")
print(f"Preferences: {prefs.value}")
progress = store.get(("users", "user_001"), "progress")
print(f"Learning progress: {progress.value}")
# Search data: search(namespace)
all_user_data = store.search(("users", "user_001"))
print(f"\n"All user data ({len(all_user_data)} items):")
for item in all_user_data:
print(f" {item.key}: {item.value}")
# Delete data: delete(namespace, key)
store.delete(("users", "user_001"), "preferences")
deleted = store.get(("users", "user_001"), "preferences")
print(f"\n"After deletion: {deleted}") # None
store = InMemoryStore()
# Write data: put(namespace, key, value)
# namespace is a tuple, key is a string, value is a dictionary
store.put(
("users", "user_001"), # Namespace
"preferences", # Key
{ # Value
"theme": "dark",
"language": "zh-CN",
"level": "Beginner",
}
)
store.put(
("users", "user_001"),
"progress",
{
"completed_courses": ["HTML Basics", "Python Basics"],
"total_hours": 35,
}
)
# Read data: get(namespace, key)
prefs = store.get(("users", "user_001"), "preferences")
print(f"Preferences: {prefs.value}")
progress = store.get(("users", "user_001"), "progress")
print(f"Learning progress: {progress.value}")
# Search data: search(namespace)
all_user_data = store.search(("users", "user_001"))
print(f"\n"All user data ({len(all_user_data)} items):")
for item in all_user_data:
print(f" {item.key}: {item.value}")
# Delete data: delete(namespace, key)
store.delete(("users", "user_001"), "preferences")
deleted = store.get(("users", "user_001"), "preferences")
print(f"\n"After deletion: {deleted}") # None
Output:
Preferences: {'theme': 'dark', 'language': 'zh-CN', 'level': '入门'}
Learning progress: {'completed_courses': ['HTML 基础', 'Python 基础'], 'total_hours': 35}
用户的所有数据 (2 项):
preferences: {'theme': 'dark', 'language': 'zh-CN', 'level': '入门'}
progress: {'completed_courses': ['HTML 基础', 'Python 基础'], 'total_hours': 35}
After deletion: None
Using Store in an Agent
Pass Store to create_agent(), and all tools in the Agent can access it via InjectedStore:
Example
from dotenv import load_dotenv
load_dotenv()
from typing import Annotated
from langgraph.store.base import BaseStore
from langgraph.store.memory import InMemoryStore
from langchain.tools import tool, InjectedStore
from langchain.agents import create_agent
from langchain.chat_models import init_chat_model
from langchain.messages import HumanMessage
# Create a Store and preset data
store = InMemoryStore()
store.put(("example", "courses"), "catalog", {
"Python3 Basics Tutorial": {"price": "Free", "hours": 20, "level": "Beginner"},
"Python Data Analysis": {"price": "Member", "hours": 30, "level": "Advanced"},
"Java Object-Oriented": {"price": "Free", "hours": 25, "level": "Advanced"},
})
store.put(("example", "users"), "user_vip_001", {
"name": "Xiao Ming",
"membership": "VIP",
"joined": "2024-01-15",
})
@tool
def query_course_info(
course_name: str,
store: Annotated[BaseStore, InjectedStore()],
) -> str:
"""Query detailed information about courses on the EXAMPLE tutorial site.
Args:
course_name: Course name
"""
item = store.get(("example", "courses"), "catalog")
catalog = item.value if item else {}
if course_name in catalog:
info = catalog[course_name]
return (
f"《{course_name}》 - Price: {info['price']},"
f"Duration: {info['hours']} hours, Difficulty: {info['level']}"
)
return f"Course 《{course_name}》 not found"
@tool
def get_user_membership(
user_id: str,
store: Annotated[BaseStore, InjectedStore()],
) -> str:
"""Query user membership information.
Args:
user_id: User ID
"""
item = store.get(("example", "users"), user_id)
if item is None:
return f"User {user_id} not found"
user = item.value
return (
f"User {user['name']}, {user['membership']} member,"
f"Registration date {user['joined']}"
)
model = init_chat_model("deepseek:deepseek-v4-flash", temperature=0)
agent = create_agent(
model=model,
tools=[query_course_info, get_user_membership],
store=store,
system_prompt="You are the course consultant for EXAMPLE.",
)
# Query course information (data comes from Store)
result = agent.invoke({
"messages": [HumanMessage(content="How much does the Python3 Basics Tutorial cost?")]
})
print(f"Query course: {result['messages'][-1].content}")
# Query user information (data comes from Store)
result = agent.invoke({
"messages": [HumanMessage(content="Help me look up information for user user_vip_001")]
})
print(f"Query user: {result['messages'][-1].content}")
load_dotenv()
from typing import Annotated
from langgraph.store.base import BaseStore
from langgraph.store.memory import InMemoryStore
from langchain.tools import tool, InjectedStore
from langchain.agents import create_agent
from langchain.chat_models import init_chat_model
from langchain.messages import HumanMessage
# Create a Store and preset data
store = InMemoryStore()
store.put(("example", "courses"), "catalog", {
"Python3 Basics Tutorial": {"price": "Free", "hours": 20, "level": "Beginner"},
"Python Data Analysis": {"price": "Member", "hours": 30, "level": "Advanced"},
"Java Object-Oriented": {"price": "Free", "hours": 25, "level": "Advanced"},
})
store.put(("example", "users"), "user_vip_001", {
"name": "Xiao Ming",
"membership": "VIP",
"joined": "2024-01-15",
})
@tool
def query_course_info(
course_name: str,
store: Annotated[BaseStore, InjectedStore()],
) -> str:
"""Query detailed information about courses on the EXAMPLE tutorial site.
Args:
course_name: Course name
"""
item = store.get(("example", "courses"), "catalog")
catalog = item.value if item else {}
if course_name in catalog:
info = catalog[course_name]
return (
f"《{course_name}》 - Price: {info['price']},"
f"Duration: {info['hours']} hours, Difficulty: {info['level']}"
)
return f"Course 《{course_name}》 not found"
@tool
def get_user_membership(
user_id: str,
store: Annotated[BaseStore, InjectedStore()],
) -> str:
"""Query user membership information.
Args:
user_id: User ID
"""
item = store.get(("example", "users"), user_id)
if item is None:
return f"User {user_id} not found"
user = item.value
return (
f"User {user['name']}, {user['membership']} member,"
f"Registration date {user['joined']}"
)
model = init_chat_model("deepseek:deepseek-v4-flash", temperature=0)
agent = create_agent(
model=model,
tools=[query_course_info, get_user_membership],
store=store,
system_prompt="You are the course consultant for EXAMPLE.",
)
# Query course information (data comes from Store)
result = agent.invoke({
"messages": [HumanMessage(content="How much does the Python3 Basics Tutorial cost?")]
})
print(f"Query course: {result['messages'][-1].content}")
# Query user information (data comes from Store)
result = agent.invoke({
"messages": [HumanMessage(content="Help me look up information for user user_vip_001")]
})
print(f"Query user: {result['messages'][-1].content}")
Output:
Course query: 《Python3 基础教程》是免费的，学习时长约20小时，难度为入门级别。 User query: 用户小明是 VIP 会员，注册日期为 2024年1月15日。
Store Persistence
InMemoryStore data is lost after a program restart. For production environments, you can use persistent solutions such as PostgresStore:
Example
# Development stage
from langgraph.store.memory import InMemoryStore
store = InMemoryStore()
# Production environment (requires PostgreSQL)
# from langgraph.store.postgres import PostgresStore
# store = PostgresStore.from_conn_string("postgresql://...")
from langgraph.store.memory import InMemoryStore
store = InMemoryStore()
# Production environment (requires PostgreSQL)
# from langgraph.store.postgres import PostgresStore
# store = PostgresStore.from_conn_string("postgresql://...")
Store Usage Recommendations
| Scenario | namespace example | key example | Description |
|---|---|---|---|
| User preferences | ("users", user_id) | preferences | Theme, language, notification settings |
| Learning progress | ("users", user_id) | progress | Completed courses, study duration |
| Knowledge base | ("kb", collection) | doc_id | Documents, FAQ, product information |
| Conversation summaries | ("sessions", thread_id) | summary | Summaries of long conversations, for use in scenarios beyond Checkpointer |
Other extensionsCheckpointer is responsible for "where the conversation is," and Store is responsible for "who the user is, what they know, and what they like." Only by using both together can you build an intelligent Agent with persistent memory.