lakebase-setup

To configure persistent memory for LangGraph agents, you provision Lakebase as the memory backend, initialize the required database schemas, and grant service-principal permissions to enable short-term and long-term memory storage across sessions.

Setting up Lakebase for agent memory requires configuring your Databricks environment, provisioning the Lakebase instance, initializing the public and agent_server schemas, and granting the necessary service-principal permissions for deployment.

Yes, LangGraph persistent memory works across sessions by using Lakebase, which provides the underlying Postgres-based storage layer to retain both short-term memory and long-term memory between agent server interactions.

The public and agent_server schemas in Lakebase manage different memory lifecycles. Initializing these schemas structures the database to support both short-term memory within active LangGraph templates and long-term memory across sessions.

Lakebase deployment fails to save agent memories when service-principal permissions are missing. You must explicitly grant deployment permissions to the service principal after initializing the database to ensure the agent memory layer functions correctly.

Provisioning Lakebase involves a seven-step workflow: configuring dependencies, provisioning the Lakebase instance, setting up the environment, initializing database schemas, and granting service-principal permissions for a working memory layer.