GitHub logo

GitHub

Free

SWE-agent takes a GitHub issue and tries to automatically fix it, using your LM of choice.

FreeFree tier
Type
Open Source

About GitHub

SWE-agent is an open-source framework that allows language models like GPT-4o or Claude Sonnet 4 to autonomously interact with tools to fix bugs in real GitHub repositories, find cybersecurity vulnerabilities, or perform custom software engineering tasks. Developed by researchers at Princeton University and Stanford University, it achieved state-of-the-art results on the SWE-bench benchmark among open-source projects. The agent operates through a configurable YAML-driven system, supporting free-flowing generalization and being designed for research extensibility. SWE-agent also includes a specialized mode (EnIGMA) for offensive cybersecurity challenges. Note: The project has been superseded by the simpler mini-SWE-agent, which matches its performance.

Key Features

Autonomous bug fixing on real GitHub repositories using any language model
Configurable via a single YAML file for full control over agent behavior
Free-flowing, generalizable tool use leaving maximal agency to the LM
Specialized mode (EnIGMA) for offensive cybersecurity (capture the flag) challenges
State-of-the-art performance on SWE-bench among open-source projects (as of Feb 2025)
Designed for research: simple, hackable architecture with thorough documentation

Pros & Cons

Pros
  • Achieves state-of-the-art results on SWE-bench among open-source solutions
  • Open source and free to use with any compatible language model
  • Highly configurable through a single YAML file
  • Designed with research principles: extensible, hackable, and well documented
  • Supports both software engineering and cybersecurity use cases
  • Backed by academic institutions (Princeton and Stanford)
Cons
  • Current development has shifted to the simpler mini-SWE-agent, which may reduce future updates for the original SWE-agent
  • Requires users to bring their own language model and API keys, increasing setup complexity
  • May be overkill for simple tasks; mini-SWE-agent recommended for most users
  • Not a standalone service; requires command-line and GitHub proficiency

Best For

Automatically fixing bugs and issues in open-source GitHub repositoriesFinding and patching cybersecurity vulnerabilities (including CTF challenges)Competitive coding challengesCustom automation tasks that require tool-using language models

FAQ

What language models does SWE-agent support?
SWE-agent supports any language model of your choice, with examples including GPT-4o and Claude Sonnet 4.
Is SWE-agent free to use?
SWE-agent is open-source and free to use. However, you may incur costs from your chosen language model provider.
What is the relationship between SWE-agent and mini-SWE-agent?
Mini-SWE-agent is a simpler version that matches SWE-agent's performance and is now the main focus of development. The original SWE-agent is still available but no longer actively improved.