langgraph-persistence

You can save LangGraph state and conversation history by using checkpointing and a key-value store to persist graph executions. This enables durable execution by maintaining thread-specific past states for your application.

Time travel in LangGraph checkpointing is the ability to replay, fork, or update application states from historical checkpoints. It allows developers to access past states for a given thread ID and modify execution paths retroactively.

Yes, LangGraph persistence supports saving and loading graph states using various backends including PostgreSQL, SQLite, and in-memory storage. This allows you to choose the appropriate database for your state management and checkpointing needs.

You add long-term memory to a LangGraph application by utilizing a Store for cross-thread memory. This mechanism stores user preferences or facts across different threads, enabling persistent context beyond a single conversation history.

Subgraph persistence mode controls persistence behavior for subgraphs to manage interrupts and multi-turn memory. This scoping allows you to define specific state saving and execution boundaries within complex graph structures.

You need state persistence for LangGraph execution when your application requires durable runs, conversation history, or time travel. Checkpointing solves this by saving and loading graph states so you can resume or review past executions reliably.