SWE-agent is an open-source autonomous software engineering agent harness built and maintained by researchers from Princeton University and Stanford University. Developed in tandem with the SWE-bench evaluation framework, it enables language models to autonomously interact with repositories, investigate bugs, execute terminal commands, and produce patches. Operating via a command-line interface, SWE-agent couples frontier language models with an Agent-Computer Interface (ACI) designed specifically for agentic code navigation and editing. Day-to-day runs can be triggered for single repository issues or across entire dataset batches. It supports proprietary models including Claude 3.7 Sonnet and GPT-4o, as well as open-weights models, with the entire agent setup—prompts, models, tools, and history processors—configured via YAML files. Unlike interactive pair-programming assistants, SWE-agent is built for autonomous execution, academic research, and benchmarking. It provides dedicated trajectory logging, a Trajectory Inspector for post-run analysis, multimodal support for issue images, and customizable tool bundles. Note that the project maintainers have shifted primary active development to mini-swe-agent, maintaining SWE-agent in maintenance mode for benchmark reproducibility. SWE-agent is open-source under the MIT license and freely available for self-hosted execution, with operational costs tied directly to the chosen LLM provider API usage.
Tags: agent harness, ai agent, cli, coding agent, developer tools, OpenSource