langgraph-persistence

LangGraph state persistence is managed using checkpointers that save and load graph states. This enables isolated thread management, allowing concurrent conversations to maintain separate execution states using in-memory, SQLite, or PostgreSQL databases.

LangGraph checkpointing supports time travel by allowing you to browse checkpoint history. You can replay past execution states and fork from previous checkpoints to explore different execution paths within your graph workflow.

LangGraph provides a long-term memory store that facilitates cross-thread memory. This allows you to maintain shared state across conversations, retaining user preferences and facts beyond the lifecycle of isolated thread execution.

LangGraph persistence provides subgraph checkpointer scoping to control persistence behavior for subgraphs. This manages multi-turn memory and handles parallel execution within isolated subgraph states, supporting complex workflow interruptions.

LangGraph state persistence requires the LangGraph framework alongside relevant database drivers like sqlite3 for SQLite or psycopg2 for PostgreSQL. These dependencies enable durable execution and permanent state storage across graph runs.

Setting up basic in-memory checkpointing in LangGraph involves using the persistence configuration command. This initializes a basic graph with thread-based state persistence for temporary, non-durable execution state management.