langgraph-persistence

To persist LangGraph state across conversations, configure a checkpointer backend like InMemorySaver, SqliteSaver, or PostgresSaver, and assign a thread_id to isolate conversation histories. This enables durable execution by checkpointing graph state across super-steps.

Sharing memory across LangGraph threads requires store integration alongside your checkpointer. While thread_id isolates individual conversation histories, the store layer handles cross-thread memory for long-term user preferences and facts.

LangGraph state management supports time travel through checkpointed states. You can inspect, replay, or fork previous checkpoints to manage conversations by accessing the history saved by your configured checkpointer backend.

For LangGraph persistence in production, PostgresSaver is recommended. InMemorySaver suits testing, SqliteSaver fits development, and PostgresSaver handles distributed deployments and durable checkpointing at scale.

Yes, a thread_id is always required for LangGraph state checkpointing. It isolates conversation histories and links per-thread conversations, ensuring state persistence is correctly scoped across steps and calls.