langgraph-persistence

To save and resume LangGraph state, compile your graph with a checkpointer and invoke it using a thread_id configuration. This persists the graph state after each step, allowing workflows to resume from their last checkpoint across threads.

LangGraph checkpointing is the mechanism of saving graph state after each execution step. You need it to enable memory, time travel, and fault tolerance across long-running workflows that require durable state management.

Yes, LangGraph persistence supports multiple built-in checkpointers including PostgresSaver, SqliteSaver, MemorySaver, and CosmosDBSaver. You can use these to manage per-thread checkpoints and cross-thread memory sharing.

Cross-thread memory sharing in LangGraph is enabled by using memory stores alongside thread-scoped checkpointers. This allows production workflows to share state across different threads while maintaining fault tolerance.

Yes, LangGraph persistence provides time travel and fault tolerance by saving graph state after each step. This allows you to restart workflows from any previous checkpoint using a thread_id based configuration.

MemorySaver stores LangGraph checkpoints in memory, making it suitable for development and testing but less ideal for production workflows. For production durability, PostgresSaver, SqliteSaver, or CosmosDBSaver are recommended to ensure state persists across restarts.