Towards Computer-Using Personal Agents

Computer-Using Agents (CUA) enable users to automate increasingly complex tasks through interaction with graphical interfaces such as web browsers. However, many such tasks require access to personal data, which raises concerns around control, interoperability, and trust. We propose Computer-Using Personal Agents (CUPA): agents that extend CUAs by leveraging a structured, user-controlled Personal Knowledge Graph (PKG). The PKG serves as an external, semantic repository of the user's private data, enabling CUPAs to reason over, personalise, and automate tasks while respecting access policies and evolving user preferences. PKGs support interoperability with external data sources and other agents, facilitate policy-aware data exchange, and allow for richer, privacy-preserving automation. CUPAs not only provide users with greater control over their data but also open new opportunities for collaborative, agent-mediated task coordination across users.