SWE-agent

SWE-agent is an open source project that lets a language model work through real GitHub issues on its own. It gives the model a set of tools and lets it decide what to read, change and test in a code repository.

You point SWE-agent at a repository and an issue. It then runs a loop in which the language model looks at the code, edits files and checks the result. Researchers from Princeton University and Stanford University build it, and it can also be used for other custom tasks, such as searching for security weaknesses. The project site says that most new development has moved to a smaller successor called mini-swe-agent, so newcomers should compare the two.

• Works on real GitHub issues and repositories
• Lets the model choose its own tools and steps
• Behavior is set in one YAML configuration file
• Accepts images from issues with vision-capable models
• Simple code that is easy to read and change

Researchers who study coding agents can use it as a test bed. Developers who want to see how a language model handles an issue, or who want a hackable base for their own agent experiments, will also benefit.

The code is released under the MIT license, so you can use and modify it at no license cost. You bring your own language model, so the cost of running it depends on the provider you choose. Start with the installation guide and the hello world tutorial in the documentation.