SWE-agent

What is SWE-agent?

SWE-agent lets a language model of your choice, such as GPT-4o or Claude Sonnet 4, use tools autonomously. It can resolve issues in real GitHub repositories, look for cybersecurity vulnerabilities or carry out custom tasks. It gives the model simple commands and feedback formats, called an Agent-Computer Interface, to browse a repository, view, edit and run code. The project is configured through a single YAML file, runs from the sweagent command line, and saves a trajectory file for each run so you can inspect every step. It supports local models and multimodal inputs, and a related EnIGMA version targets capture-the-flag security challenges. Researchers at Princeton University and Stanford University build and maintain it, and it is aimed at AI researchers and engineers who want a simple, hackable baseline. The documentation now recommends mini-swe-agent, which supersedes it, and SWE-agent is in maintenance-only mode. It is free and open source; you pay for the model API you use.

Use cases

• Resolve GitHub issues automatically
• Benchmark language models on SWE-bench
• Find security vulnerabilities in code
• Run custom agent tasks with your own tools
• Replay and inspect agent trajectories
• Research agent-computer interfaces

Who it's for

• AI researchers
• Software engineers
• ML engineers
• Security researchers

Why use SWE-agent

• Model-agnostic: swap between hosted and local models by editing a config file.
• Full trajectories make it straightforward to debug what the agent did.
• Documented, simple and hackable, which suits experiments and research.
• A clear pointer to mini-swe-agent shows where active development has moved.

SWE-agent features

• Works with GPT-4o, Claude and local models
• Agent-Computer Interface for browsing and editing code
• Single YAML file configuration
• sweagent run and run-batch commands
• Trajectory files recording every action
• Custom tools support
• Multimodal support for images in issues
• Prompt caching guidance for Anthropic Claude
• Command-line and web inspectors
• EnIGMA mode for cybersecurity challenges

SWE-agent integrations

• GitHub
• Docker
• Claude
• GPT-4o
• litellm

How to get started with SWE-agent

1. Install SWE-agent and set up Docker plus an API key for your chosen language model.
2. Point sweagent run at a GitHub issue or repository, or use run-batch for many issues.
3. The model uses tools in a container to explore, edit and test the code in an agent loop.
4. Review the output trajectory and generated patch with the inspector commands.
5. Adjust the YAML config, tools or demonstrations and rerun.

SWE-agent limitations

• Now in maintenance-only mode, superseded by mini-swe-agent
• Depends on Docker containers for your environment
• Model usage costs are paid to your model provider

SWE-agent pricing

SWE-agent is open source, so you can use and self-host it for free.

Prices can change and may vary by country or exclude tax. Check the official pricing page.

SWE-agent alternatives

The main SWE-agent alternatives listed on Debuly are OpenHands, Aider, Devin, Claude Code, Bodega One Code and Gito. The table compares pricing and platforms; order is A–Z.

1. Ai

   Aider

   Open-source AI pair programming tool that runs in your terminal and edits code in your local git repository using the LLM you choose.

   CodingDeveloper platforms
   Open sourceVisit Aider (official site)
2. 

   Bodega One Code

   Free local-first AI coding agent and full IDE that runs parallel background agents on your machine with any local or cloud model you bring.

   CodingDeveloper platformsAutomation and agents
   FreeVisit Bodega One Code (official site)
3. 

   Claude Code

   Anthropic's coding agent that fixes bugs, writes tests and runs migrations from your terminal, IDE, desktop app, web or Slack.

   CodingAutomation and agentsDeveloper platforms
   PaidVisit Claude Code (official site)
4. 

   Devin

   Cognition's AI software engineer, with a desktop app, CLI and cloud agents that take coding tasks through to pull requests.

   CodingAutomation and agentsDeveloper platforms
   FreemiumVisit Devin (official site)
5. Gi

   Gito

   Open-source AI code reviewer that checks GitHub pull requests and local changes with the language model provider you choose.

   CodingDeveloper platformsAutomation and agents
   Open sourceVisit Gito (official site)
6. 

   OpenHands

   Open-source AI coding agent platform with a free Agent Canvas, a Teams plan, and self-hosted Enterprise control

   CodingAutomation and agentsDeveloper platforms
   FreemiumVisit OpenHands (official site)

SWE-agent FAQ

Sources & verification

Facts on this page come from the official sources below, each with the date we read it. Written by Debuly; the vendor did not review this page.

Image credits

• Logo: none collected yet, so a monogram is shown.
• Cover: none collected yet, so a generated tile is shown.

Spot an error? Report an issue with this listing. We also remove a logo or image on the owner's request.