SWE-Agent is an open-source research project from the Princeton NLP Group that turns language models into software engineering agents capable of autonomously resolving real-world GitHub issues. Its key innovation is the Agent-Computer Interface (ACI), a purpose-built command set and feedback loop that lets LLMs navigate codebases and make edits far more effectively than raw bash access. Published at NeurIPS 2024, it achieved state-of-the-art results on SWE-bench. SWE-Agent 2.0 added improvements for context retention in long-horizon coding tasks. Key features: - Agent-Computer Interface (ACI) optimized for LLM codebase navigation and editing - Supports any LLM backend (GPT-4, Claude, open-source models) - Autonomous GitHub issue resolution end-to-end - State-of-the-art SWE-bench performance (NeurIPS 2024) - Usable for offensive cybersecurity and competitive coding challenges - Fully open-source, self-hostable
Free and open-source (Apache 2.0). No hosted paid tier.