Databricks adds Agent Memory in beta, so agents keep facts between sessions

Databricks has released Agent Memory in beta on Agent Bricks, its platform for building agents. It records each conversation, keeps a separate store of facts, preferences and workflows worth remembering, and runs a background process that removes duplicates and replaces facts that newer conversations contradict; developers can read, export and delete what an agent keeps about each user. Databricks says that in its internal benchmarks this design tripled agents' pass rate on real enterprise tasks while using 31% fewer tokens. In one example it gives, a related follow-up question took 2 tool calls and 41 seconds, down from 11 tool calls and nearly six minutes. These are Databricks' own results.