SWE-agent is a Python project that uses a language model to autonomously fix GitHub issues, with additional modes for cybersecurity challenges and custom tasks. It provides a YAML-configurable framework and documentation, maintained by researchers from Princeton and Stanford.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
SWE-agent enables your language model of choice (e.g. GPT-4o or Claude Sonnet 4) to autonomously use tools to fix issues in real GitHub repositories, find cybersecurity vulnerabilities, or perform any custom task. The project is built and maintained by researchers from Princeton University and Stanford University.
How it works
The README describes a workflow where a LM is controlled through a configurable setup to interact with GitHub issues, potentially fixing them automatically. It also includes a dedicated mode (EnIGMA) for offensive cybersecurity challenges. The configuration is governed by a single YAML file and the project emphasizes free-flowing agency for the LM.
Getting started
Read our documentation for installation and usage:
- Installation: https://swe-agent.com/latest/installation/source/
- Hello world from the command line: https://swe-agent.com/latest/usage/hello_world/
- Benchmarking on SWE-bench: https://swe-agent.com/latest/usage/batch_mode/
- Frequently Asked Questions: https://swe-agent.com/latest/faq/
Try SWE-agent in your browser:
👉 Open in GitHub Codespaces: https://codespaces.new/SWE-agent/SWE-agent
Recent releases
Latest releases include:
- v1.1.0 (2025-05-22): 10s of thousands of training trajectories; notes reference SWE-smith project and SWE-agent 1.0.x lineage.
- v1.0.1 (2025-02-28): SOTA on SWE-Bench Full; claims state of the art on the full set of SWE-bench issues.
- v1.0.0 (2025-02-13): Initial SWE-agent release.
- v0.7.0 (2024-09-25): EnIGMA mode announced for cybersecurity; SWE-agent is SOTA on offensive cybersecurity.
- v0.6.1 (2024-06-20): Patch release addressing issues from speed improvements in v0.7.0.
Traction
Stars: 19994
Behind the repo
SWE-agent is built and maintained by researchers from Princeton University and Stanford University. Related projects include mini-SWE-agent and SWE-bench ecosystem, but the README emphasizes SWE-agent as the primary research project.
Caveats
License: MIT. The repository notes that Mini-SWE-agent supersedes SWE-agent in practice, with guidance to prefer mini-SWE-agent for similar performance with simpler implementation. No additional licensing or age notes beyond the MIT license.