ReflectWorld is live world memory for AI agents. It turns camera streams into persistent visual experience: what happened, who or what appeared, what changed, and what an agent should remember next.</p>\n","updatedAt":"2026-07-21T03:34:50.752Z","author":{"_id":"669530098b906774beec4a2f","avatarUrl":"/avatars/ea9a9aa6e0b61de184f13ccbf0a45e37.svg","fullname":"can lee","name":"Canlee","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8647986650466919},"editors":["Canlee"],"editorAvatarUrls":["/avatars/ea9a9aa6e0b61de184f13ccbf0a45e37.svg"],"reactions":[],"isReport":false},"replies":[{"id":"6a5f044a0232a5972b6ab0e4","author":{"_id":"64a51d673badec105d75fe57","avatarUrl":"/avatars/c0166a66b468de07b101ec4bc690707b.svg","fullname":"Bo Qiao（Not In Microsoft）","name":"bo3333","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2026-07-21T05:31:54.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Nice paper! Looking forward seeing it be used in industry soon!","html":"<p>Nice paper! Looking forward seeing it be used in industry soon!</p>\n","updatedAt":"2026-07-21T05:31:54.453Z","author":{"_id":"64a51d673badec105d75fe57","avatarUrl":"/avatars/c0166a66b468de07b101ec4bc690707b.svg","fullname":"Bo Qiao（Not In Microsoft）","name":"bo3333","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.69344562292099},"editors":["bo3333"],"editorAvatarUrls":["/avatars/c0166a66b468de07b101ec4bc690707b.svg"],"reactions":[],"isReport":false,"parentCommentId":"6a5ee8dab7dbb879d8fbede5"}}]},{"id":"6a5f05b0e11418004402d6f0","author":{"_id":"6751ab1b1f1b0ff35792c236","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6751ab1b1f1b0ff35792c236/gWA7YCGDWem5i8epl5WHy.png","fullname":"abhinandan sendhav","name":"Abhi4545","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2026-07-21T05:37:52.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Really interesting approach to entity-centric memory instead of frame-based storage. Curious about the memory scalability aspect — for continuous real-world deployment (e.g., smart glasses running for months), how does the system handle memory growth for highly repetitive daily entities? Does the hierarchical compression fully offset long-term storage costs, or is there a practical upper bound before retrieval latency degrades?","html":"<p>Really interesting approach to entity-centric memory instead of frame-based storage. Curious about the memory scalability aspect — for continuous real-world deployment (e.g., smart glasses running for months), how does the system handle memory growth for highly repetitive daily entities? Does the hierarchical compression fully offset long-term storage costs, or is there a practical upper bound before retrieval latency degrades?</p>\n","updatedAt":"2026-07-21T05:37:52.562Z","author":{"_id":"6751ab1b1f1b0ff35792c236","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6751ab1b1f1b0ff35792c236/gWA7YCGDWem5i8epl5WHy.png","fullname":"abhinandan sendhav","name":"Abhi4545","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8883064985275269},"editors":["Abhi4545"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/6751ab1b1f1b0ff35792c236/gWA7YCGDWem5i8epl5WHy.png"],"reactions":[],"isReport":false},"replies":[{"id":"6a5f2135c1ce970a64dc35d0","author":{"_id":"6423dfe84e0ba2457af11787","avatarUrl":"/avatars/ad4debb13bdd1e4a54dd8aa3ef928d93.svg","fullname":"MA","name":"SylvainMA","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2026-07-21T07:35:17.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Thanks for the thoughtful question and for your interest in our work!\nFor repetitive daily entities, new observations are associated with the same recognized entity, and its semantic memory is periodically consolidated and updated rather than creating a separate profile each time. However, the corresponding fine-grained episodic records still accumulate, so hierarchical memory reduces retrieval and context costs but does not make lifetime storage constant.\nAt query time, the system retrieves a bounded set of relevant memories from an indexed database instead of scanning the full history. We therefore expect retrieval to remain practical for a typical single-camera deployment, but we have not yet conducted a months-scale stress test and cannot report a universal latency limit. For indefinite deployment, we plan to introduce importance-aware retention that removes redundant, low-value episodes while preserving important evidence. We consider this an important direction for future evaluation.","html":"<p>Thanks for the thoughtful question and for your interest in our work!<br>For repetitive daily entities, new observations are associated with the same recognized entity, and its semantic memory is periodically consolidated and updated rather than creating a separate profile each time. However, the corresponding fine-grained episodic records still accumulate, so hierarchical memory reduces retrieval and context costs but does not make lifetime storage constant.<br>At query time, the system retrieves a bounded set of relevant memories from an indexed database instead of scanning the full history. We therefore expect retrieval to remain practical for a typical single-camera deployment, but we have not yet conducted a months-scale stress test and cannot report a universal latency limit. For indefinite deployment, we plan to introduce importance-aware retention that removes redundant, low-value episodes while preserving important evidence. We consider this an important direction for future evaluation.</p>\n","updatedAt":"2026-07-21T07:35:17.584Z","author":{"_id":"6423dfe84e0ba2457af11787","avatarUrl":"/avatars/ad4debb13bdd1e4a54dd8aa3ef928d93.svg","fullname":"MA","name":"SylvainMA","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9062804579734802},"editors":["SylvainMA"],"editorAvatarUrls":["/avatars/ad4debb13bdd1e4a54dd8aa3ef928d93.svg"],"reactions":[],"isReport":false,"parentCommentId":"6a5f05b0e11418004402d6f0"}}]}],"primaryEmailConfirmed":false,"paper":{"id":"2607.09759","authors":[{"_id":"6a55aecfa9d74d6e65bbd3b5","name":"Xiaokang Ma","hidden":false},{"_id":"6a55aecfa9d74d6e65bbd3b6","name":"Yifan Sun","hidden":false},{"_id":"6a55aecfa9d74d6e65bbd3b7","name":"Zhihong Jin","hidden":false},{"_id":"6a55aecfa9d74d6e65bbd3b8","name":"Jie Gu","hidden":false},{"_id":"6a55aecfa9d74d6e65bbd3b9","name":"Yudong Luo","hidden":false},{"_id":"6a55aecfa9d74d6e65bbd3ba","name":"Shenyi Shao","hidden":false},{"_id":"6a55aecfa9d74d6e65bbd3bb","name":"Chu Tang","hidden":false},{"_id":"6a55aecfa9d74d6e65bbd3bc","name":"Jingmin Chen","hidden":false},{"_id":"6a55aecfa9d74d6e65bbd3bd","name":"Li Pu","hidden":false}],"publishedAt":"2026-07-14T00:00:00.000Z","submittedOnDailyAt":"2026-07-21T00:00:00.000Z","title":"ReflectWorld-MM: An Entity-Oriented Multimodal Memory System for Open-Ended Video Streams","submittedOnDailyBy":{"_id":"669530098b906774beec4a2f","avatarUrl":"/avatars/ea9a9aa6e0b61de184f13ccbf0a45e37.svg","isPro":false,"fullname":"can lee","user":"Canlee","type":"user","name":"Canlee"},"summary":"Building assistants that can continually watch the world, remember what they see, and reason over their accumulated experience is a long-standing goal, and recently multimodal agents equipped with long-term memory over video streams have attracted increasing interest. Unfortunately, existing systems either keep their memory inside the model context or in a flat feature store, and organize it around frames rather than around the persistent entities a stream is really about, which confines them to bounded videos and weakens their ability to track who and what reappears over time. In this paper, we propose ReflectWorld-MM, an entity-oriented multimodal memory system for open-ended video streams. It consists of three parts. The first is a perception front-end that turns an audiovisual stream into entity-resolved observations under a bounded short-term memory. The second is a hierarchical long-term memory, grounded in human memory theory, that couples a multi-scale episodic memory, an evolving entity-centric semantic memory, and a procedural memory. The third is a complete realization, built for real-world operation, that ingests arbitrary streams and plugs into off-the-shelf assistants. Across six long-video and lifelong-memory benchmarks, ReflectWorld-MM achieves the best accuracy on all six, outperforming strong memory agents and a frontier model.","upvotes":19,"discussionId":"6a55aecfa9d74d6e65bbd3be","projectPage":"https://github.com/addxai/ReflectWorld","githubRepo":"https://github.com/addxai/ReflectWorld","githubRepoAddedBy":"user","githubStars":68},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"6423dfe84e0ba2457af11787","avatarUrl":"/avatars/ad4debb13bdd1e4a54dd8aa3ef928d93.svg","isPro":false,"fullname":"MA","user":"SylvainMA","type":"user"},{"_id":"655c8302aa23904bd1fb7ae7","avatarUrl":"/avatars/4337e5a96f862773952430afe265c257.svg","isPro":false,"fullname":"Jie Gu","user":"m2408gj","type":"user"},{"_id":"67dab6fc70e82bd911eb08af","avatarUrl":"/avatars/f9d53cd0fe8846c608cbae1d005bb703.svg","isPro":false,"fullname":"Wu Xinhe","user":"lovepeace123","type":"user"},{"_id":"6401b2a93e3d0f2745ab705a","avatarUrl":"/avatars/d45a45eee8ccb0fbed2de04f1ed046ae.svg","isPro":false,"fullname":"caoxh","user":"swave2023","type":"user"},{"_id":"690daf437628f9d72083337c","avatarUrl":"/avatars/ddcd870cc83103a1c4f6304dbd90b367.svg","isPro":false,"fullname":"RuyiWang","user":"KudouZala","type":"user"},{"_id":"64afa6f3a6599cc217a5971f","avatarUrl":"/avatars/508fdd783a821e657a109f6bf8fb57a0.svg","isPro":false,"fullname":"laura","user":"laura7794","type":"user"},{"_id":"64362992f3b08e267d9e5cc4","avatarUrl":"/avatars/e23383da2acac4d7aa17af7fd68b8c68.svg","isPro":false,"fullname":"white-loub","user":"white-loub","type":"user"},{"_id":"64218b51e61513ec1c63d6d4","avatarUrl":"/avatars/2a14f872a391737b3d862788ab9f99df.svg","isPro":false,"fullname":"Jingmin Chen ","user":"Jingmin","type":"user"},{"_id":"64999d4b75cc04ed6d20c5cc","avatarUrl":"/avatars/18c0e79aac850bd2c32cf32ba1f3c46d.svg","isPro":false,"fullname":"Rui-Qi Wang","user":"Maox","type":"user"},{"_id":"6a59e6c32169cbe1519dfac2","avatarUrl":"/avatars/e86befae929463668c3f87eb3714be40.svg","isPro":false,"fullname":"penghui","user":"penghuiw","type":"user"},{"_id":"6772524ed6f92f429bd343a3","avatarUrl":"/avatars/211e0c4641b2d048b0136d7cdeef2483.svg","isPro":false,"fullname":"Zuming Huang","user":"zuminghuang","type":"user"},{"_id":"67091f4d99a18734c1364f1b","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/2l4cm4a41GO3eRS_CnOsp.png","isPro":false,"fullname":"AGI","user":"NextNextNext","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"query":{}}">