Agent state and memory

Agents need persistent storage to maintain context across turns and sessions. Lakebase Autoscaling provides a fully managed Postgres backend for storing agent state and memory, integrating natively with Databricks authentication and scaling automatically with your workload.

Short-term vs. long-term memory

Short-term memory	Long-term memory
Captures context within a single conversation session using thread IDs and checkpointing. Lets agents answer follow-up questions with awareness of earlier turns.	Extracts and stores key insights across multiple conversations. Enables personalized responses based on past interactions. Builds a user knowledge base that improves over time.

You can implement either or both memory types in the same agent.

Deployment options

Lakebase-backed agent memory is supported on two Databricks deployment targets:

Databricks Apps: Deploy agents as interactive applications with short-term or long-term memory, using LangGraph checkpointers or the OpenAI Agents SDK. Databricks handles authentication between the app and Lakebase automatically.

See Self-managed agent memory (Lakebase).

Model Serving: Deploy agents to Model Serving endpoints with Lakebase-backed checkpoints. Supports LangGraph time travel to resume or fork conversations from any checkpoint. See Agent memory (Model Serving).

Implementation

For full setup instructions, app templates, and notebook examples, see:

• Self-managed agent memory (Lakebase)

• Agent memory (Model Serving)

Additional resources

• Using Lakebase with Databricks Apps
• Autoscaling
• Author an agent and deploy it on Databricks Apps