SWE-agent

SWE-agent is an open-source autonomous software engineering agent from Princeton NLP that interfaces with GitHub repositories via a custom agent-computer interface (ACI). You point it at a GitHub issue, it clones the repo, explores the codebase, writes a fix, and submits a patch. It scored highly on the SWE-bench benchmark — outperforming many commercial agents when it launched.

Best for AI researchers studying agent frameworks, engineers experimenting with autonomous coding pipelines, and open-source maintainers curious about automating issue resolution. The architecture and agent-computer interface design are educational — many commercial agentic coding products borrowed ideas from SWE-agent's design.

Where it's not production-ready: pass rate on real-world issues is lower than benchmarks suggest once you move beyond isolated, well-defined bugs. It requires careful setup, API credits, and human oversight to avoid bad patches. Not a replacement for Devin or Claude-based agents in a commercial context, but the best open-source reference implementation for autonomous coding agents available.