# SWE-agent Autonomous coding agent that takes GitHub issues and fixes them using LLMs, achieving state-of-the-art results on SWE-bench. ## Summary SWE-agent is an open-source autonomous coding agent that takes GitHub issues as input and produces pull request fixes using large language models. It achieves state-of-the-art results on the SWE-bench benchmark and provides a configurable agent-computer interface (ACI) for optimizing how LLMs interact with development environments. ## Guide SWE-agent is one of the most important projects in the autonomous coding agent space. It demonstrated that LLMs can autonomously fix real-world GitHub issues, and its agent-computer interface design influenced how many subsequent agents structure their tool interactions. ### What it is SWE-agent is an open-source autonomous coding agent that takes GitHub issues as input and produces pull request fixes. It uses an agent-computer interface (ACI) that defines how the LLM interacts with the development environment — file navigation, code search, editing, and testing. The agent operates autonomously, requiring no human intervention from issue to PR. ### Why it matters The SWE-bench benchmark, which SWE-agent helped popularize, tests agents on real-world GitHub issues from popular Python repositories. SWE-agent's state-of-the-art results on this benchmark demonstrated that autonomous coding is viable, not just a research curiosity. Its ACI design also showed that tool interface design is a critical factor in agent performance. ### How it works Given a GitHub issue, SWE-agent clones the repository, navigates the codebase to understand the problem, identifies the relevant files, makes edits, runs tests to verify the fix, and creates a pull request. The ACI provides optimized tools for each step — search commands that work well with LLM context windows, editing interfaces that minimize errors, and testing workflows that catch regressions. ## Use Cases - Automated bug fix pipelines: Feed GitHub issues to SWE-agent and let it attempt fixes automatically, submitting PRs for human review. - Agent benchmarking research: Use SWE-agent as a baseline for benchmarking new agent architectures, tool designs, and model capabilities. - Agent-computer interface research: Study how different tool designs affect agent performance by modifying SWE-agent's ACI and measuring results on SWE-bench. ## Alternatives - Use Claude Code for interactive coding sessions vs Claude Code: SWE-agent is autonomous and issue-driven. Claude Code is interactive and conversational, better suited for iterative development workflows. - Use Aider for model-agnostic pair programming vs Aider: SWE-agent is designed for autonomous operation. Aider is designed for interactive pair programming with model flexibility. ### Getting Started - Read the documentation: https://swe-agent.com/docs - Inspect the repository: https://github.com/SWE-agent/SWE-agent ### FAQ - Is SWE-agent open source? - Yes. SWE-agent is released under the MIT license, making it fully open source for inspection, modification, and redistribution. - Who should use SWE-agent? - Researchers studying autonomous coding agents, teams building automated bug-fixing pipelines, and developers interested in agent-computer interface design. - How does SWE-agent compare to other coding agents? - SWE-agent is unique in its focus on autonomous issue fixing and benchmark performance. Unlike interactive tools like Aider or Claude Code, it's designed to operate without human intervention. ## Why It Matters SWE-agent matters because it was one of the first projects to demonstrate that LLMs can autonomously fix real-world GitHub issues at scale. Its agent-computer interface design influenced how many subsequent coding agents structure their tool interactions, and its benchmark results set a standard for the field. ## Best For - Researchers benchmarking autonomous coding agent capabilities - Teams studying how LLMs interact with real-world development environments - Developers building autonomous issue-fixing pipelines for GitHub repositories ## Not For - Developers who need interactive, conversational coding assistance - Teams looking for a production-ready CI/CD integration for automated bug fixing ## What It Actually Does - Autonomous issue fixing: SWE-agent takes a GitHub issue as input and autonomously produces a fix, from understanding the problem to writing the code and creating a pull request. - Why it matters: This is the end-to-end vision for coding agents: describe a problem and get a fix, with no human intervention required. - Agent-computer interface design: SWE-agent's ACI defines how the agent interacts with the development environment — file navigation, search, editing, and testing — optimized for LLM capabilities. - Why it matters: The ACI design influenced many subsequent coding agents and demonstrates that how an agent interacts with tools matters as much as the model itself. - SWE-bench benchmark leader: Achieves state-of-the-art results on the SWE-bench benchmark, which tests agents on real-world GitHub issues from popular Python repositories. - Why it matters: Benchmark results provide objective evidence of capability and help researchers compare different agent approaches. ## Typical Use Cases - Autonomous bug fixing research: Use SWE-agent to study how well LLMs can understand bugs, navigate codebases, and produce correct fixes without human guidance. - Agent design experimentation: Experiment with different ACI configurations to understand how tool design affects agent performance on coding tasks. - Issue triage and fixing pipelines: Build pipelines that automatically attempt to fix incoming GitHub issues and submit PRs for human review. ## How It Compares - Choose SWE-agent for autonomous issue fixing vs interactive coding agents: SWE-agent is designed for autonomous operation: give it an issue and get a fix. Interactive tools like Aider or Claude Code are better for conversational coding sessions. ## Command Line ### Install SWE-agent Install via pip, then configure your GitHub token and model provider to start running autonomous issue fixes. ```bash pip install swe-agent ``` ## Facts - Category: agents - Resource type: agent - Open source: yes - License: MIT - Last verified: 2026-05-27 - GitHub repo: SWE-agent/SWE-agent - GitHub stars: 19300 ## Capabilities - workflow-orchestration ## Structured Use Case Tags - developer-workflow ## Getting Started - Open the GitHub repository: https://github.com/SWE-agent/SWE-agent - Read the documentation: https://swe-agent.com/docs - Visit the project website: https://swe-agent.com ## Links - GitHub: https://github.com/SWE-agent/SWE-agent - Homepage: https://swe-agent.com - Docs: https://swe-agent.com/docs ## Structured Outputs - JSON: https://www.openagent.bot/agents/swe-agent.json - Markdown: https://www.openagent.bot/agents/swe-agent.md - Canonical: https://www.openagent.bot/agents/swe-agent