OS-Copilot is a new framework introduced to build digital agents capable of interfacing with comprehensive elements in an operating system. The framework is used to create FRIDAY, a self-improving agent for automating general computer tasks. FRIDAY outperforms previous methods by 35% on GAIA, a general AI assistants benchmark, and shows strong generalization to unseen applications via skills learned from previous tasks. The work provides infrastructure and insights for future research towards more capable and general-purpose computer agents.

 

Publication date: 12 Feb 2024
Project Page: https://os-copilot.github.io/
Paper: https://arxiv.org/pdf/2402.07456