langgraph-persistence

You can persist LangGraph agent state to PostgreSQL by configuring a PostgresSaver backend, which enables durable checkpointing and thread-scoped conversation history retention across multi-turn execution cycles.

Time travel debugging in LangGraph allows developers to browse execution history and replay past states. Checkpointing works by saving graph execution state to backends like SQLite or PostgreSQL for later forking and error recovery.

You store cross-thread user preferences and long-term memory for LangGraph agents by utilizing the Store feature, which maintains persistent facts across different conversation sessions and independent chat threads.

You can use either SQLite or PostgreSQL for LangGraph checkpointing. SQLite works for local development, while PostgreSQL serves as a production-grade backend for durable state persistence and long-term memory management.

To fork LangGraph execution paths and test different agent decisions, browse the checkpointed execution history to select a past state, then replay or fork the execution from that specific step for testing and error recovery.