Agent Memory is now available in Beta on Agent Bricks, Databricksโ agent developer platform. It gives agents durable sessions and long-term memory for facts, preferences, and workflows, so they can provide more personalized experiences while reducing repeated work across conversations.
Key highlights
- Persistent context across sessions: Managed sessions preserve conversation history, while managed memory stores durable facts such as user preferences, past decisions, and domain knowledge.
- Memory that improves over time: Background memory pipelines distill conversations into durable memories, merge overlapping entries, and update facts when newer conversations contradict them.
- Flexible recall and control: Agents can recall or save memories through tools, or applications can retrieve relevant memories before a run. Developers can read, export, and delete stored entries and steer what a pipeline retains.
- Built for production agents: Session and memory stores are managed on Lakebase and work with frameworks including LangGraph, the OpenAI Agents SDK, and other Python frameworks.
- Part of the Agent Bricks platform: Combine memory with agent hosting, the Agent Bricks CLI, Databricks Sandbox, Unity Gateway, MCP tools, tracing, and evaluation.
๐ Read the full post here ๐