SWE-agent is a software engineering tool that employs language models, such as GPT-4, to autonomously resolve bugs and issues within real GitHub repositories. It achieves this by utilizing an Agent-Computer Interface (ACI) to streamline interactions between the language model and the repository's codebase, enabling the model to browse, view, edit, and execute files more effectively.