Databricks launched Agent Memory in beta for Agent Bricks, giving enterprise AI agents the ability to retain useful facts, preferences, corrections and successful workflows across interactions.

The capability is based on Databricks Research’s MemCurrent architecture and runs on Lakebase. In internal benchmarks, Databricks said the memory system tripled agent pass rates while using 31% fewer tokens, potentially improving both agent quality and operating efficiency.

Source: databricks.com