Deprecated: The each() function is deprecated. This message will be suppressed on further calls in /home/zhenxiangba/zhenxiangba.com/public_html/phproxy-improved-master/index.php on line 456 Paper page - MIRIX: Multi-Agent Memory System for LLM-Based Agents
\n\t\t\n\t\n\t\n\t\t🚀 MIRIX: The AI Memory Revolution\n\t\n\n
🌟 What is MIRIX? ✅ 6 Specialized Memories: Core, Episodic, Semantic, Procedural, Resource, and Knowledge Vault — no other system covers so much. ✅ Multi-Agent Architecture: Eight intelligent agents work together to store, retrieve, and reason over everything you share. ✅ Multi-Modal Power: MIRIX ingests and compresses high-resolution screenshots, not just text — achieving 35% higher accuracy with 99.9% less storage. ✅ State-of-the-Art Results: On LOCOMO, MIRIX reached 85.4% accuracy, beating all known baselines.
\n","updatedAt":"2025-07-11T01:07:19.848Z","author":{"_id":"63234809155b0e2c44f354d6","avatarUrl":"/avatars/60d38f8f0e12363f3f5e0388e635d7b6.svg","fullname":"Yu Wang","name":"YuWangX","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":12,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7672719955444336},"editors":["YuWangX"],"editorAvatarUrls":["/avatars/60d38f8f0e12363f3f5e0388e635d7b6.svg"],"reactions":[{"reaction":"🔥","users":["YuWangX","prnvpwr2612","Eryk-Chmielewski","ai-hyz"],"count":4}],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2507.07957","authors":[{"_id":"68706310c8391850d60977de","user":{"_id":"63234809155b0e2c44f354d6","avatarUrl":"/avatars/60d38f8f0e12363f3f5e0388e635d7b6.svg","isPro":false,"fullname":"Yu Wang","user":"YuWangX","type":"user"},"name":"Yu Wang","status":"claimed_verified","statusLastChangedAt":"2025-07-11T08:02:31.566Z","hidden":false},{"_id":"68706310c8391850d60977df","name":"Xi Chen","hidden":false}],"publishedAt":"2025-07-10T17:40:11.000Z","submittedOnDailyAt":"2025-07-10T23:37:19.840Z","title":"MIRIX: Multi-Agent Memory System for LLM-Based Agents","submittedOnDailyBy":{"_id":"63234809155b0e2c44f354d6","avatarUrl":"/avatars/60d38f8f0e12363f3f5e0388e635d7b6.svg","isPro":false,"fullname":"Yu Wang","user":"YuWangX","type":"user"},"summary":"Although memory capabilities of AI agents are gaining increasing attention,\nexisting solutions remain fundamentally limited. Most rely on flat, narrowly\nscoped memory components, constraining their ability to personalize, abstract,\nand reliably recall user-specific information over time. To this end, we\nintroduce MIRIX, a modular, multi-agent memory system that redefines the future\nof AI memory by solving the field's most critical challenge: enabling language\nmodels to truly remember. Unlike prior approaches, MIRIX transcends text to\nembrace rich visual and multimodal experiences, making memory genuinely useful\nin real-world scenarios. MIRIX consists of six distinct, carefully structured\nmemory types: Core, Episodic, Semantic, Procedural, Resource Memory, and\nKnowledge Vault, coupled with a multi-agent framework that dynamically controls\nand coordinates updates and retrieval. This design enables agents to persist,\nreason over, and accurately retrieve diverse, long-term user data at scale. We\nvalidate MIRIX in two demanding settings. First, on ScreenshotVQA, a\nchallenging multimodal benchmark comprising nearly 20,000 high-resolution\ncomputer screenshots per sequence, requiring deep contextual understanding and\nwhere no existing memory systems can be applied, MIRIX achieves 35% higher\naccuracy than the RAG baseline while reducing storage requirements by 99.9%.\nSecond, on LOCOMO, a long-form conversation benchmark with single-modal textual\ninput, MIRIX attains state-of-the-art performance of 85.4%, far surpassing\nexisting baselines. These results show that MIRIX sets a new performance\nstandard for memory-augmented LLM agents. To allow users to experience our\nmemory system, we provide a packaged application powered by MIRIX. It monitors\nthe screen in real time, builds a personalized memory base, and offers\nintuitive visualization and secure local storage to ensure privacy.","upvotes":80,"discussionId":"68706311c8391850d60977e0","projectPage":"https://mirix.io/","githubRepo":"https://github.com/Mirix-AI/MIRIX","githubRepoAddedBy":"user","ai_summary":"MIRIX, a modular multi-agent memory system, enhances language models' memory capabilities by integrating diverse memory types and a dynamic framework, achieving superior performance in multimodal and long-form conversation benchmarks.","ai_keywords":["modular","multi-agent memory system","Core","Episodic","Semantic","Procedural","Resource Memory","Knowledge Vault","ScreenshotVQA","LOCOMO","long-form conversation benchmark","memory-augmented LLM agents"],"githubStars":3535},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"63234809155b0e2c44f354d6","avatarUrl":"/avatars/60d38f8f0e12363f3f5e0388e635d7b6.svg","isPro":false,"fullname":"Yu Wang","user":"YuWangX","type":"user"},{"_id":"661d91f812216c0003e8dc47","avatarUrl":"/avatars/82c857b9619bd3a96642a8a84aba2b58.svg","isPro":false,"fullname":"xichen","user":"cybergate","type":"user"},{"_id":"647bc5022d27d3541df04a91","avatarUrl":"/avatars/e062092036da6c38f700362c0c5437c0.svg","isPro":false,"fullname":"Tianxin Wei","user":"tianxinwei","type":"user"},{"_id":"67e88c1d048f3bf1dd03468a","avatarUrl":"/avatars/ae9f276b41a4a78e89b74833f33d38b2.svg","isPro":false,"fullname":"Shijun Li","user":"xiwenchao","type":"user"},{"_id":"686647f3e2cca4bc45818704","avatarUrl":"/avatars/1efeff78892223b6e5592d4ef994f7c3.svg","isPro":false,"fullname":"YUANZHE HU","user":"ai-hyz","type":"user"},{"_id":"6512dbb324347ff4170e3390","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6512dbb324347ff4170e3390/_f1H4aG4t1_qllL63f0Mn.jpeg","isPro":false,"fullname":"AIThink","user":"xinsliu","type":"user"},{"_id":"64e6783a9ac2be39203cd068","avatarUrl":"/avatars/170e2743dcdb1ea0e2cc15628a8849ef.svg","isPro":false,"fullname":"Xidong Wu","user":"Xidongwu","type":"user"},{"_id":"64b868fc77ae61bcc8f69bfb","avatarUrl":"/avatars/26cb90177f8ac4106213a8c0f786497a.svg","isPro":false,"fullname":"Zeyuan Chen","user":"zec016","type":"user"},{"_id":"646323556c27a7e33b23f198","avatarUrl":"/avatars/17fe142f689ab4be3c2374d1d90393db.svg","isPro":false,"fullname":"Yunzhe Li","user":"yunzhel2","type":"user"},{"_id":"64d07c40c9d00e3847fffd5d","avatarUrl":"/avatars/93399a3e78534af5d6be1af362a80f7f.svg","isPro":true,"fullname":"Zaiyi Zheng","user":"dontpanica","type":"user"},{"_id":"672191c1d3084146fa140541","avatarUrl":"/avatars/bde6aa6db6d92a5913ce3861dcbc55b0.svg","isPro":false,"fullname":"N.A.","user":"cxkjntmngm","type":"user"},{"_id":"643060c6cb3fe707b24c53a2","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/643060c6cb3fe707b24c53a2/MIoM9hrX0vV4XRyrm-4Kz.jpeg","isPro":false,"fullname":"Zachary Novack","user":"ZacharyNovack","type":"user"}],"acceptLanguages":["*"],"dailyPaperRank":2}">
MIRIX, a modular multi-agent memory system, enhances language models' memory capabilities by integrating diverse memory types and a dynamic framework, achieving superior performance in multimodal and long-form conversation benchmarks.
AI-generated summary
Although memory capabilities of AI agents are gaining increasing attention,
existing solutions remain fundamentally limited. Most rely on flat, narrowly
scoped memory components, constraining their ability to personalize, abstract,
and reliably recall user-specific information over time. To this end, we
introduce MIRIX, a modular, multi-agent memory system that redefines the future
of AI memory by solving the field's most critical challenge: enabling language
models to truly remember. Unlike prior approaches, MIRIX transcends text to
embrace rich visual and multimodal experiences, making memory genuinely useful
in real-world scenarios. MIRIX consists of six distinct, carefully structured
memory types: Core, Episodic, Semantic, Procedural, Resource Memory, and
Knowledge Vault, coupled with a multi-agent framework that dynamically controls
and coordinates updates and retrieval. This design enables agents to persist,
reason over, and accurately retrieve diverse, long-term user data at scale. We
validate MIRIX in two demanding settings. First, on ScreenshotVQA, a
challenging multimodal benchmark comprising nearly 20,000 high-resolution
computer screenshots per sequence, requiring deep contextual understanding and
where no existing memory systems can be applied, MIRIX achieves 35% higher
accuracy than the RAG baseline while reducing storage requirements by 99.9%.
Second, on LOCOMO, a long-form conversation benchmark with single-modal textual
input, MIRIX attains state-of-the-art performance of 85.4%, far surpassing
existing baselines. These results show that MIRIX sets a new performance
standard for memory-augmented LLM agents. To allow users to experience our
memory system, we provide a packaged application powered by MIRIX. It monitors
the screen in real time, builds a personalized memory base, and offers
intuitive visualization and secure local storage to ensure privacy.
🌟 What is MIRIX? ✅ 6 Specialized Memories: Core, Episodic, Semantic, Procedural, Resource, and Knowledge Vault — no other system covers so much. ✅ Multi-Agent Architecture: Eight intelligent agents work together to store, retrieve, and reason over everything you share. ✅ Multi-Modal Power: MIRIX ingests and compresses high-resolution screenshots, not just text — achieving 35% higher accuracy with 99.9% less storage. ✅ State-of-the-Art Results: On LOCOMO, MIRIX reached 85.4% accuracy, beating all known baselines.