Skip to content

SWE-agent

FieldValue
DeveloperPrinceton NLP
LanguagePython
LicenseMIT
GitHubSWE-agent/SWE-agent
Stars15K+

SWE-agent is a research framework from Princeton NLP that enables LLMs to autonomously fix real GitHub issues. It achieved the first open-source SOTA on SWE-bench by providing LLMs with a custom Agent-Computer Interface (ACI) — specialized commands for navigating, searching, and editing codebases.

  • Agent-Computer Interface (ACI) — Custom shell commands optimized for LLM code editing
  • SWE-bench SOTA — Proven on real-world GitHub issues at scale
  • Configurable agents — YAML-based agent configuration
  • Multi-model — Works with GPT-4o, Claude Sonnet, and other models
  • Docker sandboxing — Safe execution environment for each issue

✅ Autonomous GitHub issue fixing ✅ SWE-bench evaluation and research ✅ Building agents that work on unfamiliar codebases

❌ Interactive development (use AdaL CLI or Aider)