What problem does it solve?
Provides a robust persistence layer for LangGraph to reliably save and restore graph execution state so conversations, long-running flows, and multi-turn agents can resume across invocations and process restarts.
Core Features & Use Cases
- Checkpointer selection: guidance and examples for InMemorySaver, SqliteSaver, and PostgresSaver with production recommendations.
- Thread-scoped short-term memory: use thread_id to isolate conversation histories and enable multi-turn interactions.
- Long-term Store: cross-thread key-value store for persistent user preferences and facts accessible from runtime.
- State history & time travel: inspect, replay, fork, and update past checkpoints to resume or branch execution.
- Subgraph scoping: rules for compile-time checkpointer modes (False, None, True) and namespace isolation for parallel subgraphs.
- Operational fixes and best practices: always supply thread_id, prefer PostgresSaver for production, use Overwrite to bypass reducers, and access store via runtime.
Quick Start
Invoke the graph with a persistent checkpointer and include a configurable.thread_id to enable durable conversation history.