OS-Copilot: Towards Generalist Computer Agents with Self-Improvement

Zhiyong Wu , Chengcheng Han , Zichen Ding , Zhenmin Weng , Zhoumianze Liu , Shunyu Yao , Tao Yu , Lingpeng Kong

🏛 Institutions: Shanghai AI Laboratory , East China Normal University , Princeton , HKU
📅 Date: February 12, 2024
📑 Publisher: LLMAgents @ ICLR 2024
💻 Env: Desktop Web
🔑 Keywords: framework FRIDAY self-directed learning skill accumulation OS-Copilot

TLDR

OS-Copilot is a framework for building generalist computer agents that interact with operating-system elements including the web, code terminals, files, multimedia, and third-party applications. The paper instantiates it with FRIDAY, a self-improving embodied agent that learns new application skills over time and reports a 35% improvement over prior methods on GAIA.

Open paper Report issue