agent-memory

You can persist user preferences and conversation state across LangGraph sessions by integrating memory tools with AsyncDatabricksStore for long-term Lakebase storage and AsyncCheckpointSaver for short-term thread_id-scoped state.

Short-term checkpointing maintains session continuity within a thread_id, while long-term Lakebase storage stores, searches, and deletes user memories that persist across sessions using AsyncDatabricksStore.

To configure Databricks Lakebase for agent memory, add the databricks-langchain memory dependency, set Lakebase environment variables, and wire memory_tools into your agent server to handle user_id extraction and safe JSON validation.

Yes, LangGraph checkpointing works with Databricks LangChain by using AsyncCheckpointSaver to maintain conversation history scoped by thread_id, while AsyncDatabricksStore handles long-term memory persistence across requests.

Yes, implementing long-term agent memory requires databricks-langchain memory, databricks-sdk, langchain-core, langgraph, and mlflow dependencies to enable storage, checkpointing, and tool integration functionality.

Agents forget user-specific information between conversations when they lack persistent memory; integrating memory_tools with AsyncDatabricksStore enables storing and retrieving preferences, facts, and history across separate requests.