MemCurrent
#13 in Agent FrameworksDatabricks · since Oct 20, 2026 (Blog-Beitrag / Ankündigung) · 2× · last seen Oct 22, 2026
31
Momentum
MemCurrent is a memory architecture for production AI agents, developed by Databricks Research, that separates raw session records from distilled knowledge. It combines automatic plugins, on-demand tools and a background consolidation step ("dreaming") and runs on Lakebase within Agent Bricks Memory. Performance figures come from Databricks' internal benchmarks.
Momentum trend
26.07.24.10.
Features
| Autonomy Level | Agent decides on-demand tool calls (save/recall); background pipeline consolidates memory automatically |
| Channels | Any agent framework (e.g., LangGraph, OpenAI templates); deployed as an app via Agent Bricks CLI |
| Modalities | Text-based memory content (free text) with semantic and keyword search |
| Platform | Agent Bricks (Databricks), storage on Lakebase (serverless Postgres); available on Azure Databricks and AWS |
| Price | No additional charges for Agent Memory itself during Beta; the underlying Lakebase instance is billed |
| Release Date | Oct 20, 2026 (Databricks blog announcement; Agent Memory in Beta) |
| Tool/MCP Integrations | Memory/recall tools; Databricks-managed MCP services, Unity Catalog functions, Genie |