--- title: MemCurrent: Designing Persistent Memory for Production... description: Databricks built MemCurrent, a persistent memory architecture now powering Agent Bricks Memory and Genie. It triples an agent's pass rate over having no... canonical: https://daily.dev/posts/memcurrent-designing-persistent-memory-for-production-agents-qyumd8wb3 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: MemCurrent: Designing Persistent Memory for Production Agents | daily.dev og:description: Databricks built MemCurrent, a persistent memory architecture now powering Agent Bricks Memory and Genie. It triples an agent's pass rate over having no... og:url: https://daily.dev/posts/memcurrent-designing-persistent-memory-for-production-agents-qyumd8wb3 og:image: https://api.daily.dev/og/posts/QYuMD8wB3.png og:image:alt: MemCurrent: Designing Persistent Memory for Production Agents og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # MemCurrent: Designing Persistent Memory for Production Agents **[databricks](https://daily.dev/sources/databricks)** · 12 min read · 1 upvotes · 0 comments ## Summary Databricks built MemCurrent, a persistent memory architecture now powering Agent Bricks Memory and Genie. It triples an agent's pass rate over having no memory, from 22.3% to 68.4%, while using 31% fewer tokens and running 26% faster. The system combines episodic memory (raw session traces) with derived memory (distilled knowledge files), accessed through plugins, agent tools, and an offline 'dreaming' pipeline. On open-source benchmarks it matches full-context performance using about 12 times fewer input tokens, 8.4K versus 103K. Ablations show derived memory alone reaches 66.1% pass rate, and combining offline dreaming with online tool writes beats either approach alone. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is episodic memory versus derived memory in an AI agent memory system? Episodic memory stores raw session traces in an append-only log, while derived memory holds distilled learnings from those traces as structured files in a knowledge base, topped by an auto-generated memory.md index. In Databricks' MemCurrent architecture, derived memory alone reaches a 66.1% pass rate on an internal benchmark, and adding episodic memory on top raises it slightly to 68.4%. _Developers designing agent memory layers can track architecture patterns like this on daily.dev._ ### How much do agent memory systems reduce token usage compared to stuffing full conversation history into context? MemCurrent matches full-context performance (88.1% vs 87.8% accuracy) while using about 12 times fewer input tokens per question, 8.4K versus 103K tokens. On an internal enterprise benchmark, it also triples pass rate over no memory (22.3% to 68.4%) while running 1.35x faster and 1.44x cheaper than no memory. _Teams weighing cost against accuracy for agent memory can follow benchmarks like these on daily.dev._ ### Does combining offline memory consolidation with online agent tool writes improve agent memory performance? Yes, combining both produces the highest pass rate, 66.1%, versus 63.5% for offline dreaming alone and 53.9% for online tools alone, on an internal Databricks memory benchmark. Offline dreaming handles background extraction and consolidation while online tools let explicit user requests take effect immediately. _Developers tuning memory extraction strategies for agents can compare approaches like this on daily.dev._ ## Similar posts on daily.dev - [Introducing Agent Memory on Databricks](https://daily.dev/posts/introducing-agent-memory-on-databricks-nzodjkaex) · databricks · 1 upvotes · 0 comments - [Cloudflare Announces Agent Memory, a Managed Persistent Memory Service for AI Agents](https://daily.dev/posts/cloudflare-announces-agent-memory-a-managed-persistent-memory-service-for-ai-agents-1umlrxax8) · InfoQ · 1 upvotes · 0 comments - [Shared Selective Persistent Memory for Agentic LLM Systems](https://daily.dev/posts/shared-selective-persistent-memory-for-agentic-llm-systems-hyy6iotpj) · Apple Machine Learning Research · 0 upvotes · 0 comments - [AI Agents Need Memory: Building Persistent Memory for Agentforce](https://daily.dev/posts/ai-agents-need-memory-building-persistent-memory-for-agentforce-db6orcgcy) · Salesforce Engineering · 5 upvotes · 0 comments - [I gave my local agent a real memory, and it finally stopped forgetting everything between sessions](https://daily.dev/posts/i-gave-my-local-agent-a-real-memory-and-it-finally-stopped-forgetting-everything-between-sessions-ydrcjuzoj) · XDA Developers · 0 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents), [#databricks](https://daily.dev/tags/databricks) [View this post on daily.dev](https://daily.dev/posts/memcurrent-designing-persistent-memory-for-production-agents-qyumd8wb3) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"MemCurrent: Designing Persistent Memory for Production Agents","url":"https://daily.dev/posts/memcurrent-designing-persistent-memory-for-production-agents-qyumd8wb3","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/memcurrent-designing-persistent-memory-for-production-agents-qyumd8wb3"},"datePublished":"2026-10-20T17:03:36.299Z","dateModified":"2026-10-20T17:24:54.378Z","description":"Databricks built MemCurrent, a persistent memory architecture now powering Agent Bricks Memory and Genie. It triples an agent's pass rate over having no...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/718d5253d65dae1e02867e3b626ffe60?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/718d5253d65dae1e02867e3b626ffe60?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"databricks","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"databricks","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/fa7aa720f2db4d1eba826814730482c8","url":"https://daily.dev/sources/databricks"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/memcurrent-designing-persistent-memory-for-production-agents-qyumd8wb3","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,ai-agents,databricks","timeRequired":"PT12M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"databricks","item":"https://daily.dev/sources/databricks"},{"@type":"ListItem","position":3,"name":"MemCurrent: Designing Persistent Memory for Production Agents"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/memcurrent-designing-persistent-memory-for-production-agents-qyumd8wb3#faq","mainEntity":[{"@type":"Question","name":"What is episodic memory versus derived memory in an AI agent memory system?","acceptedAnswer":{"@type":"Answer","text":"Episodic memory stores raw session traces in an append-only log, while derived memory holds distilled learnings from those traces as structured files in a knowledge base, topped by an auto-generated memory.md index. In Databricks' MemCurrent architecture, derived memory alone reaches a 66.1% pass rate on an internal benchmark, and adding episodic memory on top raises it slightly to 68.4%. Developers designing agent memory layers can track architecture patterns like this on daily.dev."}},{"@type":"Question","name":"How much do agent memory systems reduce token usage compared to stuffing full conversation history into context?","acceptedAnswer":{"@type":"Answer","text":"MemCurrent matches full-context performance (88.1% vs 87.8% accuracy) while using about 12 times fewer input tokens per question, 8.4K versus 103K tokens. On an internal enterprise benchmark, it also triples pass rate over no memory (22.3% to 68.4%) while running 1.35x faster and 1.44x cheaper than no memory. Teams weighing cost against accuracy for agent memory can follow benchmarks like these on daily.dev."}},{"@type":"Question","name":"Does combining offline memory consolidation with online agent tool writes improve agent memory performance?","acceptedAnswer":{"@type":"Answer","text":"Yes, combining both produces the highest pass rate, 66.1%, versus 63.5% for offline dreaming alone and 53.9% for online tools alone, on an internal Databricks memory benchmark. Offline dreaming handles background extraction and consolidation while online tools let explicit user requests take effect immediately. Developers tuning memory extraction strategies for agents can compare approaches like this on daily.dev."}}]} ```