SWE-agent

Free plan LinuxMacSelf-hostedWebWindows

What SWE-agent is

SWE-agent is an open-source coding agent that lets a language model use terminal and repository tools for autonomous work. It can address issues in GitHub repositories, investigate cybersecurity vulnerabilities, and perform custom tasks. Its Agent-Computer Interface supplies commands and feedback for browsing repositories, viewing and editing files, and running code. The project is designed for research and customization, with documented settings controlled through one YAML file. Users can add tools and alter the demonstration trajectory provided at the start of a run. SWE-agent supports models available through LiteLLM and local models exposed through an OpenAI-compatible API. Docker is the default runtime, while cloud code evaluation can run without Docker. Vision-capable models can process GitHub issue images under 10 MB. Installation is available from source with pip or a browser-based VS Code setup. It runs on Linux, macOS, and Windows, but the project is now maintenance-only and maintainers recommend mini-swe-agent instead.

Who it suits

It suits researchers and developers who want a configurable, hackable coding agent with terminal and repository access. Users should be prepared to supply model keys and, when needed, GitHub credentials.

What it does well

Open-source and free to use
Autonomous repository browsing, editing, and code execution
Configurable through a documented YAML file
Supports LiteLLM and local models
Runs on Linux, macOS, and Windows

What to know first

Now in maintenance-only mode
Maintainers recommend mini-swe-agent instead
Requires model credentials and sometimes GitHub credentials
Docker may limit environment choices

Verdict

SWE-agent offers a configurable research-oriented agent for repository work, vulnerability investigation, and custom tasks across major desktop systems. Its maintenance-only status and successor recommendation make mini-swe-agent the direction maintainers currently favor.

Facts from SWE-agent's own pages

Coding agent: Yes · github.com, 22 Sept 2026
Browser setup: SWE-agent can be run in a browser-based VS Code environment, where users provide language-model API keys in a .env file. · swe-agent.com, 11 Oct 2026
Browser use: The setup guide offers a browser-based VS Code environment for a quick first look and lists Node.js as a requirement for the web-based GUI. · swe-agent.com, 12 Oct 2026
Configuration: The project describes itself as configurable through a YAML file and designed to be simple and hackable for research. · swe-agent.com, 12 Oct 2026
Credentials: Users supply their own language-model keys and need a GitHub key to access private repositories. · swe-agent.com, 12 Oct 2026
Current recommendation: The maintainers recommend mini-swe-agent, which they say matches SWE-agent’s performance while being simpler. · swe-agent.com, 11 Oct 2026
Customization: Users can add custom tools and change the demonstration trajectory supplied to the agent at the start of each run. · swe-agent.com, 12 Oct 2026
Images: SWE-agent supports vision-capable models for processing images from GitHub issues, and the FAQ says images must be under 10 MB. · swe-agent.com, 12 Oct 2026
Image support: The documentation describes multimodal support for processing images from GitHub issues and says images must be under 10 MB. · swe-agent.com, 11 Oct 2026
Installation: The preferred setup is installation from source using pip. · swe-agent.com, 11 Oct 2026
Local model limits: For local models, the documentation says cost tracking cannot be used unless a custom model registry supplies cost information. · swe-agent.com, 11 Oct 2026
Models: The documentation says SWE-agent supports models served through LiteLLM and local models with an OpenAI-compatible API. · swe-agent.com, 12 Oct 2026
Model support: The documentation says SWE-agent supports models available through LiteLLM and local models served through an OpenAI-compatible API. · swe-agent.com, 11 Oct 2026
Operating systems: The FAQ says SWE-agent runs on Windows, macOS, and Linux, with Docker container availability possibly limiting environment choices. · swe-agent.com, 11 Oct 2026
Platforms: The FAQ says SWE-agent runs on Windows, macOS, and Linux, with Docker environment availability as a possible limitation. · swe-agent.com, 12 Oct 2026
Project status: The documentation says SWE-agent has been superseded by mini-swe-agent and is now in maintenance-only mode. · swe-agent.com, 12 Oct 2026
Purpose: SWE-agent uses language models to autonomously work on software engineering tasks, including fixing issues in GitHub repositories. · swe-agent.com, 12 Oct 2026
Research design: Its Agent-Computer Interface provides commands and feedback for models to browse repositories, view and edit files, and run code. · swe-agent.com, 12 Oct 2026
Research focus: The project describes itself as made for research and simple and hackable by design. · swe-agent.com, 11 Oct 2026
Run records: The FAQ says `.traj` output files contain complete records of SWE-agent's thought process and actions. · swe-agent.com, 12 Oct 2026
Runtime: Docker is the default backend; the documentation says cloud code evaluation is an option for users who do not want to use Docker. · swe-agent.com, 12 Oct 2026
Status: SWE-agent has been superseded by mini-swe-agent and is now in maintenance-only mode. · swe-agent.com, 11 Oct 2026
Support: The FAQ directs users to open GitHub issues for questions, bug reports, and feature requests, and links to Slack discussions. · swe-agent.com, 12 Oct 2026
Terminal access: Yes · github.com, 22 Sept 2026