Thanks for checking out MemSlides!</p>\n<p>MemSlides is a hierarchical memory-driven agent framework for personalized slide generation and multi-turn local revision.</p>\n<p>Links:</p>\n<ul>\n<li>Project page: <a href=\"https://memslides.github.io/\" rel=\"nofollow\">https://memslides.github.io/</a></li>\n<li>Code: <a href=\"https://github.com/huohua325/Memslides\" rel=\"nofollow\">https://github.com/huohua325/Memslides</a></li>\n<li>Demo Website: <a href=\"https://memslides.com/\" rel=\"nofollow\">https://memslides.com/</a></li>\n<li>arXiv: <a href=\"https://arxiv.org/abs/2606.17162\" rel=\"nofollow\">https://arxiv.org/abs/2606.17162</a></li>\n</ul>\n<p>The main idea is to separate persistent user profile memory, session-level working memory, and reusable tool memory, so the slide generation agent can personalize initial decks and perform reliable localized revisions across turns.</p>\n","updatedAt":"2026-06-17T03:19:17.588Z","author":{"_id":"672e158cd044f482ad10e4c6","avatarUrl":"/avatars/49fbbe6fa06842571466f820d09e3335.svg","fullname":"Ye Jin","name":"huohua325","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7387793660163879},"editors":["huohua325"],"editorAvatarUrls":["/avatars/49fbbe6fa06842571466f820d09e3335.svg"],"reactions":[],"isReport":false}},{"id":"6a38eba31d74395d0003926c","author":{"_id":"672e158cd044f482ad10e4c6","avatarUrl":"/avatars/49fbbe6fa06842571466f820d09e3335.svg","fullname":"Ye Jin","name":"huohua325","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2026-06-22T08:00:35.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"We introduce **MemSlides**, a memory-driven Slides Agent for personalized slide generation and multi-turn local revision.\n\nThe main question is: can a Slides Agent learn user preferences from multi-turn interactions, preserve session-level constraints, and revise only the intended parts of a deck?\n\nMemSlides organizes memory from two perspectives:\n- temporal scope: long-term memory vs. working memory\n- functional role: preference memory vs. tool memory\n\nThis lets the agent separate persistent user preferences, current-session constraints, and reusable tool-execution experience, instead of simply appending all history into the prompt.\n\nWe also study localized revision, where a user request is mapped to the smallest effective slide region to reduce unintended drift in the rest of the deck.\n\nCode, project page, and demo are linked below. Feedback is very welcome!","html":"<p>We introduce <strong>MemSlides</strong>, a memory-driven Slides Agent for personalized slide generation and multi-turn local revision.</p>\n<p>The main question is: can a Slides Agent learn user preferences from multi-turn interactions, preserve session-level constraints, and revise only the intended parts of a deck?</p>\n<p>MemSlides organizes memory from two perspectives:</p>\n<ul>\n<li>temporal scope: long-term memory vs. working memory</li>\n<li>functional role: preference memory vs. tool memory</li>\n</ul>\n<p>This lets the agent separate persistent user preferences, current-session constraints, and reusable tool-execution experience, instead of simply appending all history into the prompt.</p>\n<p>We also study localized revision, where a user request is mapped to the smallest effective slide region to reduce unintended drift in the rest of the deck.</p>\n<p>Code, project page, and demo are linked below. Feedback is very welcome!</p>\n","updatedAt":"2026-06-22T08:00:35.145Z","author":{"_id":"672e158cd044f482ad10e4c6","avatarUrl":"/avatars/49fbbe6fa06842571466f820d09e3335.svg","fullname":"Ye Jin","name":"huohua325","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8377514481544495},"editors":["huohua325"],"editorAvatarUrls":["/avatars/49fbbe6fa06842571466f820d09e3335.svg"],"reactions":[],"isReport":false}},{"id":"6a398e7984f507cc94ba2eb5","author":{"_id":"6960eca92f7ad9b043b5cbe0","avatarUrl":"/avatars/e68dcc7fd04f143d849d40414866e633.svg","fullname":"Noah","name":"noahml","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":0,"isUserFollowing":false},"createdAt":"2026-06-22T19:35:21.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Neat approach to slide generation. Most presentation tools struggle with maintaining consistency over multiple edits, so separating the memory into user profiles, working memory, and tool execution sounds like a much more stable way to handle revisions than just regenerating everything from scratch.\n\nI am curious, how does the system decide which parts of the deck are considered the smallest affected region during a local edit? Does it rely on the tool memory to map specific prompts to slide components?\n\nI made a podcast on it with ResearchPod, it makes it easy to get the key concepts on the go:\nhttps://researchpod.app/episode/afdb2aa6-91fc-4fbb-abf4-ce2fcf22d6ae","html":"<p>Neat approach to slide generation. Most presentation tools struggle with maintaining consistency over multiple edits, so separating the memory into user profiles, working memory, and tool execution sounds like a much more stable way to handle revisions than just regenerating everything from scratch.</p>\n<p>I am curious, how does the system decide which parts of the deck are considered the smallest affected region during a local edit? Does it rely on the tool memory to map specific prompts to slide components?</p>\n<p>I made a podcast on it with ResearchPod, it makes it easy to get the key concepts on the go:<br><a href=\"https://researchpod.app/episode/afdb2aa6-91fc-4fbb-abf4-ce2fcf22d6ae\" rel=\"nofollow\">https://researchpod.app/episode/afdb2aa6-91fc-4fbb-abf4-ce2fcf22d6ae</a></p>\n","updatedAt":"2026-06-22T19:35:21.906Z","author":{"_id":"6960eca92f7ad9b043b5cbe0","avatarUrl":"/avatars/e68dcc7fd04f143d849d40414866e633.svg","fullname":"Noah","name":"noahml","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":0,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.917995035648346},"editors":["noahml"],"editorAvatarUrls":["/avatars/e68dcc7fd04f143d849d40414866e633.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2606.17162","authors":[{"_id":"6a320c6cbc818ff14e453dc3","user":{"_id":"672e158cd044f482ad10e4c6","avatarUrl":"/avatars/49fbbe6fa06842571466f820d09e3335.svg","isPro":false,"fullname":"Ye Jin","user":"huohua325","type":"user","name":"huohua325"},"name":"Ye Jin","status":"admin_assigned","statusLastChangedAt":"2026-06-18T16:51:58.183Z","hidden":false},{"_id":"6a320c6cbc818ff14e453dc4","name":"Yangyang Xu","hidden":false},{"_id":"6a320c6cbc818ff14e453dc5","name":"Jun Zhu","hidden":false},{"_id":"6a320c6cbc818ff14e453dc6","name":"Yibo Yang","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/672e158cd044f482ad10e4c6/UTMh7VE3VzY6pPlrpcj02.png","https://cdn-uploads.huggingface.co/production/uploads/672e158cd044f482ad10e4c6/7XF3gWh_6TN4S6gVfVVgv.mp4","https://cdn-uploads.huggingface.co/production/uploads/672e158cd044f482ad10e4c6/fYnu7LeBAtG6aBH2ZOsbF.png","https://cdn-uploads.huggingface.co/production/uploads/672e158cd044f482ad10e4c6/6kRPquhWA5BfLkOi8Ibd7.png"],"publishedAt":"2026-06-15T00:00:00.000Z","submittedOnDailyAt":"2026-06-22T00:00:00.000Z","title":"MemSlides: A Hierarchical Memory Driven Agent Framework for Personalized Slide Generation with Multi-turn Local Revision","submittedOnDailyBy":{"_id":"672e158cd044f482ad10e4c6","avatarUrl":"/avatars/49fbbe6fa06842571466f820d09e3335.svg","isPro":false,"fullname":"Ye Jin","user":"huohua325","type":"user","name":"huohua325"},"summary":"Personalized presentation generation requires more than conditioning on a current prompt or template: agents must preserve stable user preferences across tasks, retain newly introduced preferences and constraints during multi-turn revision, and carry out local edits reliably. We propose MemSlides, a hierarchical memory framework for personalized presentation agents that separates long-term memory from working memory and further divides long-term memory into user profile memory and tool memory. User profile memory stores intent-conditioned profiles for round-0 personalization, working memory carries active preferences and session constraints across revision rounds, and tool memory stores reusable execution experience for reliable localized editing. MemSlides pairs this memory design with scoped slide-local revision, so targeted updates act on the smallest affected region instead of repeatedly regenerating the full deck. In controlled experiments, user profile memory improves persona-alignment judgments on a multi-persona, multi-intent profile bank, tool-memory injection improves closed-loop modify behavior in diagnostic matched-pair settings, and qualitative cases illustrate working memory's ability to carryover preferences. Taken together, these results suggest that effective personalization in presentation authoring depends on separating persistent user profiles, session-level working memory, and reusable execution experience across generation and localized revision.","upvotes":14,"discussionId":"6a320c6dbc818ff14e453dc7","projectPage":"https://memslides.github.io/","githubRepo":"https://github.com/huohua325/Memslides","githubRepoAddedBy":"user","ai_summary":"MemSlides presents a hierarchical memory framework for personalized presentation agents that separates long-term user profiles, working memory for session constraints, and tool memory for reusable execution experiences to enable stable personalization and reliable local edits across multi-turn revisions.","ai_keywords":["personalized presentation agents","hierarchical memory framework","long-term memory","working memory","user profile memory","tool memory","round-0 personalization","session constraints","localized editing","multi-turn revision"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","githubStars":10},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"672e158cd044f482ad10e4c6","avatarUrl":"/avatars/49fbbe6fa06842571466f820d09e3335.svg","isPro":false,"fullname":"Ye Jin","user":"huohua325","type":"user"},{"_id":"68dfd0bf9fdaa34002f781e6","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/bnZHbkRK4QVDVkWokJYfg.png","isPro":false,"fullname":"xinyi huang","user":"Hxywww666","type":"user"},{"_id":"679e39a596621e70c2cf3079","avatarUrl":"/avatars/1dfa9452d9ee2f09311e52fb558f7a8a.svg","isPro":false,"fullname":"Guoziyi","user":"EnderRieck","type":"user"},{"_id":"6a2da6c8ca070ee12c6e396c","avatarUrl":"/avatars/0355287dcabaa67dbc7f0b10b87451f9.svg","isPro":false,"fullname":"Joe Mama","user":"JoeMama123123123","type":"user"},{"_id":"6625385193567d7bbfe4c553","avatarUrl":"/avatars/c3529570d0ae605a18c1fd1c4b0f95d4.svg","isPro":false,"fullname":"Yibo Yang","user":"iboing","type":"user"},{"_id":"66f2ddc80dc1833d4e2313f0","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/4ztCwF-hifxaF-TLkP2bk.jpeg","isPro":false,"fullname":"zhang","user":"MrZhanggggg","type":"user"},{"_id":"69c4f2430b8e19faf9082483","avatarUrl":"/avatars/9b89764016620c675889e84b76358653.svg","isPro":false,"fullname":"zhou","user":"jack234dsaffsdf","type":"user"},{"_id":"69c4f3cb0b8e19faf9084411","avatarUrl":"/avatars/d3709c1d43709695466fe827a136eab8.svg","isPro":false,"fullname":"afdfjl","user":"jojf23423424","type":"user"},{"_id":"69c543346d459ab5301e71f9","avatarUrl":"/avatars/f2167e81a961815ce580d6a0afcbbaeb.svg","isPro":false,"fullname":"Spider","user":"feng9998","type":"user"},{"_id":"69c5459c77afe431d09281ed","avatarUrl":"/avatars/46ff3c2fbdf19083421e3e0a1ff04efa.svg","isPro":false,"fullname":"ssss","user":"Dongye3325345","type":"user"},{"_id":"69c546c877b3f9ac9d11c88c","avatarUrl":"/avatars/87915ff26efc0eec867593cfb7f83f34.svg","isPro":false,"fullname":"tang","user":"cong23cs","type":"user"},{"_id":"662df118b0d404635d7a2b46","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/662df118b0d404635d7a2b46/JtcwRctghymSkw67zvoPn.jpeg","isPro":false,"fullname":"Zhe Ren","user":"Ray368","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":2,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2606/2606.17162.md","query":{}}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2606.17162 in a model README.md to link it from this page.

Cite arxiv.org/abs/2606.17162 in a dataset README.md to link it from this page.

Add this paper to a collection to link it from this page.