Coco 提出深度个性化：桌面主动智能体的本地上下文观察方案
Deep Personalization: Continuous Context Observation for Local Proactive Agents
做主动式智能体的开发者可以看看这篇。Coco 用独立的 Observer Agent 在本地观察桌面行为，判断什么时候帮你、什么时候别打扰，介入判断的 F1 提升了 2.4-3.0 倍。数据不出设备这点对企业场景很关键，还附带 OSWorld-Persona 基准可直接用来测自己的方案。
一篇 arXiv 论文（编号 2611.13457）提出'deep personalization'闭环框架，并发布了桌面主动智能体 Coco。其核心是一个独立的 Observer Agent，持续监控桌面活动、维护紧凑的长期用户建模上下文，并通过互动、拒绝、显式请求、重复工作模式等隐式信号学习何时介入最合适。
技术上，团队开发了端侧高效运行策略，让 Observer Agent 可以在本地 AI 算力平台上工作，无需将原始活动数据上传云端。为避免收集敏感真实用户数据，他们还构建了数据合成管线，把现有 computer-use 轨迹转换成带 persona 条件干预标签的 OSWorld-Persona 基准。
评测显示，在 OSWorld-Persona 上，deep personalization 将三个模型骨干的 macro F1 从直接提示的 0.12-0.19 提升到适配后的 0.35-0.48，相对增益 2.4-3.0 倍；为期一周的用户研究中，Coco 的 macro F1 从 0.199 升至 0.515。对做主动式智能体和端侧部署的开发者来说，这篇论文同时给出了训练思路和可复用的基准。
Deep Personalization: Continuous Context Observation for Local Proactive Agents
As AI agents become increasingly capable, users must still recognize when and how to ask for help, leaving many opportunities for useful support unrealized. Proactive agents hold the promise to close this gap, but effective proactivity requires understanding users' real-time activities and personalizing intervention timing so that assistance is offered when it is likely to be helpful rather than disruptive. We introduce deep personalization, a closed-loop process that connects continuous context observation with continual agent adaptation: an agent observes user activity, selectively retains long-term context, and adapts intervention decisions from implicit feedback. We initiate deep personalization in Coco, a proactive desktop agent, by having a separate Observer Agent that monitors desktop activity, maintains compact long-term user modeling context, and learns when assistance is desirable from implicit signals such as engagement, dismissal, explicit requests, and repeated work patterns. We further develop efficient on-device strategies to enable the Observer Agent to run on local AI compute platforms without streaming raw activity data off-device. For evaluation at scale without collecting sensitive real-user activity, we introduce a data-synthesis pipeline that transforms existing computer-use trajectories into OSWorld-Persona, a benchmark with persona-conditioned intervention labels. On OSWorld-Persona, deep personalization improves macro F1 across three model backbones from 0.12-0.19 under direct prompting to 0.35-0.48 after adaptation, a 2.4-3.0x relative gain. In a week-long user study, we observe that Coco improves its macro-averaged F1 score from 0.199 to 0.515. These results suggest that on-device deep personalization can make proactive agents more timely and increasingly aligned with each user's everyday work.
This report is categorized under AITOP's "Papers" desk. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include Coco, OSWorld-Persona, 智能体, 端侧部署.
Source attribution: curated from arXiv cs.LG (@Yijia Shao, Song Bian, Junxian Guo, Jindong Jiang, Zhen Dong, Huizi Mao, Diyi Yang, Song Han).
