Checkpoints and thread_id

A checkpointer saves graph state at step boundaries. thread_id tells it which sequence of checkpoints belongs to this conversation or job.¹

Tiny example

from langgraph.checkpoint.memory import InMemorySaver

graph = builder.compile(checkpointer=InMemorySaver())
config = {"configurable": {"thread_id": "tenant-7:case-104"}}

graph.invoke({"messages": [{"role": "user", "content": "Check case 104"}]}, config)
snapshot = graph.get_state(config)

Term	Scope	Example
State	Current data	messages and case status
Checkpoint	One saved snapshot	state after policy lookup
Thread	Ordered checkpoint history	all work for case 104
Store	Data shared outside one thread	a user’s saved preference

Failure note

InMemorySaver loses data when the process stops; use a persistent checkpointer in production. Never let two tenants share a thread_id, and apply retention rules because checkpoint histories grow.

LangChain, Persistence. ↩