This paper introduces a framework for proactive contextual curation in personal information management systems (PIMS) using generative AI as a cognitive partner. The framework is operationalized through a contextual intelligence model that captures and predicts user needs across temporal, task-based, cognitive, and relational contexts. Unlike reactive recommendation systems, our approach employs large language models (LLMs) to anticipate user needs by analysing temporal, task-based, cognitive, and relational context. The system dynamically filters and prioritizes information streams to reduce decision fatigue, implementing a cognitive load-aware pipeline that adapts content delivery based on real-time behavioural signals. The work advances human-AI collaboration in information systems, offering a blueprint for adaptive interfaces that balance automation with user agency. Key contributions include the model as the central component of the framework, empirical validation of cognitive load reduction, and design principles for trustworthy proactive systems. Future directions emphasize persistent memory architectures and domain-specific applications.