langgraph-persistence

You persist LangGraph state by configuring a checkpointer backend like sqlite or PostgresSaver, defining a thread_id in the runtime config, and compiling the graph to save state across invocations.

Time-travel checkpointing in LangGraph allows you to inspect the history of a graph's execution. It works by storing state snapshots at every step, enabling you to review or revert to previous checkpoint states.

You can share memory across multiple chat threads by using a central Store. The Store maintains long-term memory and cross-thread preferences, functioning independently of isolated thread_id conversations.

LangGraph persistence supports subgraph orchestration through configurable subgraph persistence modes. You configure these modes at compile time to maintain state within complex subgraph workflows.

LangGraph supports InMemorySaver, sqlite, and PostgresSaver checkpointer backends. InMemorySaver is used for development, while sqlite and PostgresSaver provide durable execution for production scenarios.

You need a thread_id to enable thread isolation within LangGraph persistence. This scoping keeps multi-user chat conversations separated and ensures state is maintained correctly for each distinct execution thread.