Reproduce “Deep Personalization: Continuous Context Observation for Local Proactive Agents” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yijia Shao, Song Bian, Junxian Guo, Jindong Jiang, Zhen Dong, Huizi Mao, Diyi Yang, Song Han
Coco learns when to offer help by having an Observer Agent interpret screenshots, retain work context, and adapt its intervention decisions from engagement, rejection, and possible missed opportunities. On OSWorld-Persona, Coco-PE with Qwen-3.5-9B reaches macro F1 0.4832, versus 0.1861 for direct prompting, averaged over four personas. The gain also brings a higher false-alarm rate, so it does not show fewer interruptions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.