Autonomous software engineering agent that resolves real GitHub issues. Built by Princeton NLP.
SWE-agent turns LLMs into autonomous software engineers capable of fixing real bugs and implementing features. It takes a GitHub issue as input and attempts to generate a pull request that fixes it. Built on a custom Agent-Computer Interface (ACI) that enables LLMs to browse repositories, edit files, and run tests. Achieves state-of-the-art performance on the SWE-bench benchmark.