Technology

SWE-agent

An open-source AI coding agent that turns GitHub issues into pull requests by interacting with a sandboxed terminal via a machine-optimized interface.

Built by researchers at Princeton and Stanford, SWE-agent is an open-source system that transforms language models like GPT-4 and Claude 3.7 into autonomous software engineers. The core breakthrough is its Agent-Computer Interface (ACI): a specialized, machine-friendly terminal designed to prevent the AI from getting lost in massive codebases. Instead of using standard bash commands, the agent uses precise tools to search, view, and edit specific line ranges. This design helped it achieve state-of-the-art performance on the SWE-bench benchmark. Developers can run SWE-agent locally using Docker to safely resolve real-world bugs, run tests, and generate pull requests on autopilot.