SWE-agent
2 Tools
SWE-agent is an open-source research project originating at Princeton University (with contributors from Stanford University) that builds AI agents capable of autonomously resolving real-world GitHub issues by browsing, editing, and executing code via an "Agent-Computer Interface." First released as princeton-nlp/SWE-agent, the work was published at NeurIPS 2024 and the team also created SWE-bench, a widely used benchmark for evaluating coding agents. The organization, now hosted at github.com/swe-agent, maintains related tooling including mini-SWE-agent and SWE-ReX, a sandboxed execution environment for AI agents.
Tools
SWE-agent lets a language model like GPT-4o or Claude autonomously use tools to fix real GitHub issues, solve custom coding tasks, or find cybersecurity vulnerabilities via its EnIGMA mode. It is a research project from Princeton and Stanford that achieves state-of-the-art results on SWE-bench and is fully configurable through a single YAML file.
A minimal open-source AI coding agent built from roughly 100 lines of Python that only uses bash (no tool-calling APIs) yet scores over 74% on SWE-bench Verified. Created by the Princeton/Stanford team behind SWE-bench and SWE-agent, it supports local, Docker, and sandboxed execution and is used internally by companies like Meta, NVIDIA, and IBM.