BOOKMARKS: Efficient Active Storyline Memory for Role-playing</p>\n","updatedAt":"2026-05-15T01:46:47.993Z","author":{"_id":"64323dd503d81fa4d26deaf9","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64323dd503d81fa4d26deaf9/x3ES8VXEZJljxDWvFWaAf.png","fullname":"Letian Peng","name":"KomeijiForce","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":12,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.6717205047607422},"editors":["KomeijiForce"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/64323dd503d81fa4d26deaf9/x3ES8VXEZJljxDWvFWaAf.png"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.14169","authors":[{"_id":"6a067a98b1a8cbabc9f09803","name":"Letian Peng","hidden":false},{"_id":"6a067a98b1a8cbabc9f09804","name":"Ziche Liu","hidden":false},{"_id":"6a067a98b1a8cbabc9f09805","name":"Yiming Huang","hidden":false},{"_id":"6a067a98b1a8cbabc9f09806","name":"Longfei Yun","hidden":false},{"_id":"6a067a98b1a8cbabc9f09807","name":"Kun Zhou","hidden":false},{"_id":"6a067a98b1a8cbabc9f09808","name":"Yupeng Hou","hidden":false},{"_id":"6a067a98b1a8cbabc9f09809","name":"Jingbo Shang","hidden":false}],"publishedAt":"2026-05-13T00:00:00.000Z","submittedOnDailyAt":"2026-05-15T00:00:00.000Z","title":"BOOKMARKS: Efficient Active Storyline Memory for Role-playing","submittedOnDailyBy":{"_id":"64323dd503d81fa4d26deaf9","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64323dd503d81fa4d26deaf9/x3ES8VXEZJljxDWvFWaAf.png","isPro":false,"fullname":"Letian Peng","user":"KomeijiForce","type":"user","name":"KomeijiForce"},"summary":"Memory systems are critical for role-playing agents (RPAs) to maintain long-horizon consistency. However, existing RPA memory methods (e.g., profiling) mainly rely on recurrent summarization, whose compression inevitably discards important details. To address this issue, we propose a search-based memory framework called BOOKMARKS, which actively initializes, maintains, and updates task-relevant pieces of bookmarks for the current task (e.g., character acting). A bookmark is structured as the answer to a question at a specific point in the storyline. For each current task, BOOKMARKS selects reusable existing bookmarks or initializes new ones (at storyline beginning) with useful questions. These bookmarks are then synchronized to the current story point, with their answers updated accordingly, so they can be efficiently reused in future grounding rounds. Compared with recurrent summarization, BOOKMARKS offers (1) active grounding for capturing task-specific details and (2) passive updating to avoid unnecessary computation. In implementation, BOOKMARKS supports concept, behavior, and state searches, each powered by an efficient synchronization method. BOOKMARKS significantly outperforms RPA memory baselines on 85 characters from 16 artifacts, demonstrating the effectiveness of search-based memory for RPAs.","upvotes":6,"discussionId":"6a067a98b1a8cbabc9f0980a","githubRepo":"https://github.com/KomeijiForce/BOOKMARKS_Koishiday_2026","githubRepoAddedBy":"user","ai_summary":"BOOKMARKS is a search-based memory framework that improves role-playing agents by actively managing task-relevant information through structured bookmarks that capture detailed character behaviors and story elements.","ai_keywords":["role-playing agents","memory systems","recurrent summarization","search-based memory","bookmarks","task-relevant pieces","grounding","synchronization","concept search","behavior search","state search"],"githubStars":4,"organization":{"_id":"697e87d12cc19315a8497001","name":"UCSanDiego","fullname":"University of California at San Diego","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/697e8687c00f332cf492d29e/KUQpvngxP4r9oBSDZwIwZ.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"64323dd503d81fa4d26deaf9","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64323dd503d81fa4d26deaf9/x3ES8VXEZJljxDWvFWaAf.png","isPro":false,"fullname":"Letian Peng","user":"KomeijiForce","type":"user"},{"_id":"64a62c2f500beb50968e5c9c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64a62c2f500beb50968e5c9c/ydpojlCWaAOF68OIaEwM7.jpeg","isPro":false,"fullname":"Yupeng Hou","user":"hyp1231","type":"user"},{"_id":"6514599ee31c0e2e3dfb5c9c","avatarUrl":"/avatars/3c3ebd14d228c4c439da542cf8ff20a8.svg","isPro":false,"fullname":"ymh233","user":"ymh233","type":"user"},{"_id":"66f4fcbc29b10ae4c990a2e0","avatarUrl":"/avatars/5d6df3aa5792a031c28274d428c46d84.svg","isPro":false,"fullname":"Kun Zhou","user":"FrancisKunZhou","type":"user"},{"_id":"6464481e4b34d49ac754c7db","avatarUrl":"/avatars/14612002e05ddb4b455308ef92a8d128.svg","isPro":false,"fullname":"Archer Liu","user":"tREeFrOGorigami","type":"user"},{"_id":"620783f24e28382272337ba4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/620783f24e28382272337ba4/zkUveQPNiDfYjgGhuFErj.jpeg","isPro":false,"fullname":"GuoLiangTang","user":"Tommy930","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"697e87d12cc19315a8497001","name":"UCSanDiego","fullname":"University of California at San Diego","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/697e8687c00f332cf492d29e/KUQpvngxP4r9oBSDZwIwZ.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.14169.md"}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2605.14169 in a model README.md to link it from this page.

Cite arxiv.org/abs/2605.14169 in a dataset README.md to link it from this page.

Cite arxiv.org/abs/2605.14169 in a Space README.md to link it from this page.