langgraph-persistence

1<overview>
2LangGraph's persistence layer enables durable execution by checkpointing graph state:
3 
4- **Checkpointer**: Saves/loads graph state at every super-step
5- **Thread ID**: Identifies separate checkpoint sequences (conversations)
6- **Store**: Cross-thread memory for user preferences, facts
7 
8**Two memory types:**
9- **Short-term** (checkpointer): Thread-scoped conversation history
10- **Long-term** (store): Cross-thread user preferences, facts
11</overview>
12 
13<checkpointer-selection>
14 
15| Checkpointer | Use Case | Production Ready |
16|--------------|----------|------------------|
17| `InMemorySaver` | Testing, development | No |
18| `SqliteSaver` | Local development | Partial |
19| `PostgresSaver` | Production | Yes |
20 
21</checkpointer-selection>
22 
23---
24 
25## Checkpointer Setup
26 
27<ex-basic-persistence>
28<python>
29Set up a basic graph with in-memory checkpointing and thread-based state persistence.
30 
31```python
32from langgraph.checkpoint.memory import InMemorySaver
33from langgraph.graph import StateGraph, START, END
34from typing_extensions import TypedDict, Annotated
35import operator
36 
37class State(TypedDict):
38    messages: Annotated[list, operator.add]
39 
40def add_message(state: State) -> dict:
41    return {"messages": ["Bot response"]}
42 
43checkpointer = InMemorySaver()
44 
45graph = (
46    StateGraph(State)
47    .add_node("respond", add_message)
48    .add_edge(START, "respond")
49    .add_edge("respond", END)
50    .compile(checkpointer=checkpointer)  # Pass at compile time
51)
52 
53# ALWAYS provide thread_id
54…