lakebase-setup

Lakebase enables PostgreSQL-backed long-term and short-term agent memory by persisting chat history across sessions and threads. It integrates with Databricks agent templates to store user facts and conversational context reliably after deployment.

Your Databricks agent fails to connect to Lakebase with permission denied migration errors because the app's service principal lacks Postgres privileges. You must grant the service principal Postgres permissions via the provided grant workflow before deploying and starting the agent.

The best way to add PostgreSQL-backed persistence to a Databricks agent template is configuring the Lakebase resource in databricks.yml and granting the service principal Postgres permissions before deployment. This ensures reliable long-term and short-term memory across sessions.

Yes, you need matching embedding environment variables when setting up Lakebase for agent memory. Correct environment variables must align with your instance or autoscaling identifiers in databricks.yml to ensure PostgreSQL-backed persistence functions properly after deployment.

To grant a Databricks app service principal Postgres permissions for Lakebase, run the lakebase-setup skill's grant workflow before deploying your agent. This addresses the most common failure mode where agents encounter permission denied or failed to connect errors.

Yes, you can use Lakebase autoscaling instances for agent chat history persistence. The setup supports both provisioned instance models and autoscaling configurations, provided you supply the correct identifiers in databricks.yml and matching embedding environment variables.