SWE-agent
Autonomous agent that takes a GitHub issue and attempts to fix it end-to-end
View on GitHubWhat it solves
Automating first-pass bug fixes: give the agent a GitHub issue URL and get back a diff without writing any agent scaffolding yourself.
Overview
SWE-agent is a research framework from Princeton and Stanford that feeds a GitHub issue to an LLM agent and lets it browse the repository, write a patch, and run tests, all through a purpose-built Agent-Computer Interface (ACI) designed to reduce hallucination on code navigation tasks. Published at NeurIPS 2024. Also used for cybersecurity exploit tasks and competitive programming.
Key facts
- Language
- Python
- License
- MIT
- Maturity
- Experimental
- Maintainer
- SWE-agent
- Reviewed
- 2026-06-11
Where to start
The Agent class shows how SWE-agent structures its observation → thought → action loop and why the ACI design choices were made, essential reading before evaluating it on your own codebase.
sweagent/agent/agents.pyTechnologies
More in this domain
LangGraph
Stateful multi-agent orchestration for LLM applications
The state management problem in multi-agent systems: how you build agents that pass context across turns without losing the execution thread.
AutoGen
Multi-agent conversation framework from Microsoft Research
How to coordinate multiple specialized agents toward a shared goal without writing low-level orchestration logic by hand.
CrewAI
Role-based multi-agent orchestration for collaborative task execution
The boilerplate problem in multi-agent setup: defining roles, delegation rules, and task routing without writing a custom orchestration graph.