SWE-agent
Takes a GitHub issue and autonomously tries to fix it with your LLM of choice; also usable for offensive cybersecurity and competitive coding challenges (NeurIPS 2024).
- Actively maintained
- Well documented
- Permissive license
Overview
SWE-agent gives an LLM a custom agent-computer interface for navigating and editing code repositories, then points it at a real GitHub issue and lets it attempt an automatic fix -- generating a pull-request-ready patch. It originates from the academic team behind the SWE-bench benchmark, which is now widely used to measure coding-agent performance across the industry.
SWE-agent is a fit for teams and researchers who want a rigorously benchmarked, research-grade autonomous issue-resolution agent, or who want to reuse the same agent-computer interface for adjacent tasks like security research and competitive programming.
- Categories
- AI Agents & Agent Tools
- Languages
- Python
- License
- MIT
Related Projects
Cline
Apache-2.0Autonomous coding agent available as an IDE extension, CLI assistant or SDK, with a human-in-the-loop approval workflow for every file change and command.
Actively maintained
- AI Agents & Agent Tools
OpenHands
MITAI-driven software development platform where autonomous agents can modify code, run commands, browse the web, and call APIs to complete engineering tasks end to end.
Actively maintained
- AI Agents & Agent Tools
Aider
Apache-2.0AI pair programming in your terminal -- Aider edits code in your local git repo, running tests and committing changes as it goes.
Well documented
- AI Agents & Agent Tools
Ponytail
MITA coding-agent skill that enforces YAGNI and minimal, pragmatic solutions -- installs into Claude Code, Cursor, Codex and 15+ other agent platforms.
Actively maintained
- AI Agents & Agent Tools
Spotted an error? Suggest an edit on GitHub.