Memory systems
Databricks Agent Memory
Beta long-term agent memory in workspace-scoped stores backed by Lakebase, with semantic and BM25 retrieval
Commercial
What it is
Our note
Entries are partitioned by actor_id so one store can hold private per-user memory or shared group memory, and memory pipelines can distil entries out of conversation sessions. Access control is enforced at the store level rather than per entry or actor, and the feature remains beta with a 100-entry search cap and no pagination.
At a glance
- Vendor
- Databricks
- Business model
- Commercial
- Pricing
- During preview, billed for the underlying Lakebase instance that stores memory entries; no additional charge for agent memory itself pricing page ↗
- Self-hosting
- No, hosted only
- Access
- AgentKit SDK (Python), REST API /api/2.0/agents/memory-stores, Agent Bricks CLI
Links and sources
- Official sitedocs.databricks.com/aws/en/agents/agent-memory/managed-memory
- Pricingwww.databricks.com/product/pricing/lakebase
- Sourcedocs.databricks.com/aws/en/agents/agent-memory/self-managed-memory
Facts verified nov 10, 2026; added nov 10, 2026. Vendor facts come from the vendor’s own pages; benchmark claims are self-reported unless we say otherwise.