Advanced10 min

Custom Checkpointer & Store

Building your own persistence backends. BaseCheckpointSaver interface, custom Store implementations, and migration strategies.

Quick Reference

→BaseCheckpointSaver is the abstract interface for checkpointing — implement put(), get(), and list() to use any storage backend
→Built-in checkpointers include MemorySaver (dev only), SqliteSaver, PostgresSaver, and RedisSaver for production use
→Custom Store implementations extend BaseStore with put(), get(), search(), and delete() for cross-thread persistent memory
→Migration strategies: version your checkpoint schema, write forward-compatible serializers, and backfill old checkpoints lazily
→Use async checkpoint savers (AsyncPostgresSaver) in production to avoid blocking the event loop during state persistence

The BaseCheckpointSaver Interface

Three methods, any backend

BaseCheckpointSaver is the abstract class for all LangGraph checkpointers. Implement three methods — put(), get_tuple(), and list() — to use any storage backend: DynamoDB, MongoDB, S3, Cassandra, or your own database.

The abstract interface — these are the methods you implement

Building a Custom Checkpointer

Complete custom checkpointer backed by Redis

Custom Store for Cross-Thread Memory

The Store is separate from the Checkpointer. Checkpointers save per-thread state (conversation history within a single thread). Stores save cross-thread data (user preferences, learned facts, long-term memory shared across conversations). BaseStore provides put(), get(), search(), and delete().

Sign in to read this article

This is a premium article. Sign in with your Google account to continue.