langgraph-persistence

To persist LangGraph state across multiple conversation turns, compile your graph with a durable checkpointer backend and pass a thread_id via the configurable runtime config to scope and save the execution state at each super-step.

Checkpointing in LangGraph is the mechanism of saving and reloading graph execution state at each super-step. You need it to enable resumable execution, multi-turn conversational flows, and time-travel replay from checkpoint history.

To share user preferences and facts across different LangGraph threads, implement a Store for cross-thread long-term memory. This allows personalization to persist beyond a single conversation session or thread_id.

Yes, LangGraph persistence supports Postgres as a production-grade checkpointer backend. You can select an appropriate backend ranging from in-memory to Postgres when compiling your graph for durable state checkpointing.

Yes, you can perform time-travel and replay a LangGraph graph from checkpoint history. By using a persistent checkpointer, the execution state is durably saved at each super-step, allowing you to reload and resume from any historical point.

To handle subgraph persistence and interrupts in LangGraph, compile your subgraphs with a persistent checkpointer. This ensures stateful behavior and durable checkpointing for subgraph scoping during execution.