Skip to content
Get packs

SWE-agent

An AI agent that tries to fix GitHub issues automatically, using the language model you choose.

GitHub Download v1.1.0
Stars
20.7k
Forks
2.3k
Licence
MIT
Language
Python
Last update
10 days ago
Type
AI model / agent

What it is

SWE-agent takes a GitHub issue and lets a language model you pick (the README mentions GPT-4o and Claude Sonnet 4 as examples) use tools to try to fix it in a real repository. It is also described as usable for offensive cybersecurity (capture-the-flag) challenges and custom coding tasks. It is a research project from Princeton and Stanford, and its README says its main development has moved to the simpler mini-SWE-agent.

  • Point it at a GitHub issue in a repository and see whether it proposes a fix
  • Run it on coding challenges with a custom task defined in configuration
  • Run batch benchmarking on SWE-bench, as described in the docs
  • Try the EnIGMA mode for capture-the-flag cybersecurity challenges (the README points to SWE-agent 0.7 for this while EnIGMA is updated for 1.0)

Great for

  • Researchers studying autonomous coding agents or running SWE-bench experiments
  • Developers who want to experiment with an LLM attempting issue fixes in their own repositories
  • Students learning how agent-computer interfaces work, since the README describes it as simple and hackable
  • People who want a configurable agent controlled by a single YAML file

Not ideal for

  • Teams wanting the project's actively recommended tool: the README says most development has moved to mini-SWE-agent, which it recommends instead
  • Anyone needing a polished, finished product: the project is framed as research code
  • People who cannot pay for or access a language model API, since the README assumes you bring your own model
Step-by-step guide: how to use SWE-agentInstall, first run, examples, tips and FAQs
SWE-agent alternativesOpenManus vs SWE-agentOpenHands vs SWE-agentDeerFlow vs SWE-agentDevika vs SWE-agentSWE-agent vs Trae AgentOpen SWE vs SWE-agentdsh-superagent vs SWE-agent

How to get it

  1. 1

    Clone and build from source

    You'll need: Git · Python (version not stated in the README excerpt)

    Clone the repository

    git clone https://github.com/SWE-agent/SWE-agent.git
  2. 2

    Run it

    • Read the installation guide at swe-agent.com/latest/installation/source/
    • Set up your language model API access as described in the docs
    • Follow the hello-world guide at swe-agent.com/latest/usage/hello_world to run it on an issue
    • Review the README's FAQ at swe-agent.com/latest/faq/ for differences from mini-SWE-agent
  3. 3

    Get the most out of it

    • Use the single YAML configuration file to change the agent's behavior without editing code
    • Follow the batch mode docs to benchmark runs on SWE-bench
    • Consider the Codespaces option to try the project without a local setup
    • Compare with mini-SWE-agent, which the README says matches performance in a simpler design

Downloads

Source code (.zip)main branch · straight from GitHubRelease v1.1.0Release notes and sourceOfficial website & docsswe-agent.com

Strengths

  • The README reports state-of-the-art results on SWE-bench among open-source projects
  • Configuration is handled by a single YAML file and the README says it is fully documented
  • MIT licensed with a public docs site, Slack, and a Codespaces option to try it in a browser

Trade-offs

  • The README says mini-SWE-agent has superseded it and is recommended instead going forward
  • The EnIGMA cybersecurity mode currently points users to older version 0.7 rather than 1.0
  • Results depend on the language model and API you supply; the README does not state costs or hardware needs

How it compares

ProjectStarsTypeLevelUpdatedBest for
SWE-agent (this)20.7kAI model / agentintermediate10 days agoResearchers studying autonomous coding agents or running SWE-bench experiments
OpenHands92.5kAppintermediate4 days agoDevelopers who want coding agents running on a server while their laptop is off
ponytail182kAI model / agentbeginner8 days agoDevelopers who use Claude Code or similar AI agents and find the generated code too long
deer-flow84.9kAI model / agentadvanced4 days agoDevelopers who want to experiment with multi-agent workflows on their own infrastructure
open-swe11.2kAI model / agentadvanced4 days agoEngineering teams that want a self-hosted coding agent they can inspect and adapt
ECC295kDev toolintermediate4 days agoDevelopers who already use Claude Code and want a shared workflow for planning, testing and review

Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.