← All updates Databricks Update One-Pager Light version
databricks Platform · Updated Jun 23, 2026

Managed Agent Memory

Enhance AI agents with long-term memory capabilities.

Beta Open docs ↗release note ↗
What it does

Managed agent memory enables AI agents to retain long-term memory across conversations, with Databricks managing the infrastructure and isolating memories by scope. Memory is stored in Unity Catalog memory stores, which are governed as Unity Catalog securables and accessed via the Unity Catalog REST API. This allows agents to remember user preferences and past interactions, improving their accuracy and efficiency over time.

Why it matters

This feature significantly enhances the user experience by allowing AI agents to provide personalized interactions based on historical context. Businesses can leverage this capability to improve customer support, increase user engagement, and streamline workflows, ultimately leading to higher productivity and satisfaction.

Key capabilities
Long-Term Memory
Agents can remember user preferences and past decisions across conversations.
Unity Catalog Governance
Memory is secured and governed through Unity Catalog, ensuring compliance and access control.
Scope Isolation
Memory entries are isolated by user or organizational scope, maintaining privacy and relevance.
API Access
Memory stores and entries are accessible via the Unity Catalog REST API for integration.
Prerequisites
A Databricks workspace with Unity Catalog enabled.
The CREATE MEMORY STORE privilege on the parent schema.
Recommended use cases
Customer support agents can remember user preferences for personalized service.AI assistants can retain context for ongoing projects across multiple sessions.Sales agents can track customer interactions and preferences for better engagement.
Limitations · remember
!Memory entries provide long-term memory only; short-term memory must be managed separately.
!Memory stores and entries can only be created and managed through the Unity Catalog REST API.
Architecture
Databricks Workspace → Unity Catalog → Memory Store → AI Agent
Get started
1
Enable Unity Catalog
Ensure Unity Catalog is enabled in your Databricks workspace.
2
Set Permissions
Grant the CREATE MEMORY STORE privilege on the parent schema.
3
Create Memory Store
Use the Unity Catalog REST API to create a memory store for your agent.
4
Add Memory Entries
Store user preferences and context as memory entries in the memory store.
5
Integrate with Agent
Connect the memory store to your agent's conversation state using the OpenAI-compatible client.
Key takeaway
Managed agent memory empowers AI agents with the ability to retain and utilize long-term memory, enhancing user interactions and operational efficiency.
2026-06-23-managed-agent-memory-is-in-beta Generated by the Databricks Updates Agent · browse all updates