Stacks/Data / ML platform
Databricks MemCurrent memory for Genie
Databricks built MemCurrent, the memory system behind Genie and Agent Bricks Memory. Agents read and write episodic and file-based derived memory through a plugin and tools, while an offline dreaming pipeline consolidates it. On an internal benchmark from real Genie sessions it tripled pass rate over no memory while using fewer tokens and running faster.
- Use case
- Persistent memory for production data agents; powers Agent Bricks Memory and Genie, where users ask questions about their enterprise data
- Category
- Data / ML platform
- Team size
- not stated
- Orchestrator / framework
- Memory is wired in three ways: a plugin that logs each step and loads context at session start, tools the agent calls on demand (search_memory, fetch_memory, add_memory, update_memory, delete_memory), and an offline 'dreaming' pipeline that consolidates memory in the background.
- Memory
- Two types in a production database: append-only episodic memory (raw session traces) and derived memory (distilled learnings stored as files in a table that simulates a hierarchical file system, with an auto-generated memory.md index injected at session start). Dreaming splits oversized files, merges duplicates, and resolves conflicts.
- Evals
- LoCoMo, LongMemEval, and ConvoMem, plus an Internal Memory Benchmark built from 12K+ real Genie sessions (3K mined candidates, an adversarial judge checking 7 criteria, LLM-judge rubrics). MemCurrent tripled pass rate over no memory (22.3% to 68.4%) with 31% fewer tokens and 26% faster runs; on open benchmarks it matched full context (88.1% vs 87.8%) at about 12x fewer input tokens (8.4K vs 103K).
- Observability
- not stated
- Cost (stated in writeup)
- Not stated
- Cost per output (stated)
- Not stated in the writeup
- WhatWeRun estimate
WhatWeRun estimate
Not enough info to estimate
Token counts are benchmark-only (8.4K input tokens per question on open benchmarks); there's no production per-session token figure or volume for Genie.
Reviewed Nov 13, 2026
- Models
- Observability tools
- not stated
- Other tools
- Mem0, PropMem, OpenClaw memory (compared baselines)
- Source
- https://www.databricks.com/blog/memcurrent-designing-persistent-memory-production-agents
The Databricks AI Research Team · Published Oct 20, 2026 · Last verified Nov 13, 2026
Related
More like this
Pinterest's Analytics Agent and its tuning loop
An internal Analytics Agent that searches Pinterest's data ecosystem (tables, dashboards, metrics) to answer questions, plus an automated 'loop engineering' process that improves how reliably it follows required steps
Netflix Pensive AI failure diagnosis
AI error extraction and on-demand root-cause debugging for failed data and ML workloads on Netflix's data platform (Maestro, Spark, Trino, Druid, Titus), where roughly 100,000 workloads fail on peak days
engineering.fb.comMeta's thin orchestrator for data pipeline agents
Automating data column onboarding across Meta recommendation-system pipelines, spanning three code repositories, three programming languages, and up to six pipeline stages