SWE-agent
by Princeton NLP
SWE-agent is an open-source system from Princeton University's NLP group that lets language models inspect repositories, edit code, and run commands to resolve GitHub issues through a purpose-built agent-computer interface.
Why Abandoned
The official GitHub repository remains publicly available and documents the current SWE-agent project and releases. No verified acquisition or shutdown announcement was found; lifecycle confidence is therefore high but not absolute.
🩺 Health Signals
- HTTP Status
- 200
- Response
- —
- SSL
- —
- Confidence
- 94%
- Domain expires
- —
- Last content change
- —
- Last tweet
- —
- API status
- —
📅 Timeline
SWE-agent announced
Princeton NLP publicly introduced SWE-agent as an open-source system for using language models to fix issues in real GitHub repositories.
Initial code released
The project's source code and setup documentation were published on GitHub under an MIT license.
Research paper submitted
The SWE-agent paper was submitted to arXiv, describing the Agent-Computer Interface and evaluation on the SWE-bench software-engineering benchmark.
SWE-agent 1.0 released
The project announced a major rewrite focused on modularity, configurability, and support for additional agent workflows.
🔄 Alternatives to SWE-agent
Cursor AI
🟢ActiveAI-first code editor built on VS Code with deep AI integration for code generation and editing.
Devin (Cognition)
🟢ActiveCognition unveiled Devin in March 2024 as 'the first AI software engineer', capable of taking a Linear ticket and shipping a pull request end-to-end. Independent reviewers (notably Carl Brown's analysis) showed the launch demo was carefully edited and that real-world success rates were modest. Cognition kept iterating and acquired Windsurf in July 2025 to give Devin a real IDE surface. The company's valuation reached $4B+ in 2025 funding rounds.
GitHub Copilot
🟢ActiveAI pair programmer that suggests code completions in your IDE, powered by OpenAI Codex.
Replit Agent
🟢ActiveReplit Agent is an AI software-building tool within Replit that turns natural-language requests into working applications. It can plan projects, write code, configure infrastructure, test changes, and deploy apps from the browser.
🪦 Other dead AI Agents tools
Sweep
🟠Possibly DeadSweep was an AI coding agent that converted GitHub issues and bug reports into proposed code changes and pull requests. Its original product was later discontinued as the team shifted to an AI-assisted JetBrains coding plugin.
Ultimate
🟡Acquired / MergedUltimate was a customer-support automation platform that used conversational AI to resolve requests across chat, email, and social channels. Zendesk acquired the company in 2024 and integrated its technology into Zendesk AI.
Hyperlight
🟡Acquired / MergedHyperlight was an AI-agent startup focused on automating customer-service workflows for businesses. Public reporting indicates that the company was acquired by Cresta in 2024 to strengthen Cresta's AI capabilities.
Sana Agents
🟡Acquired / MergedSana Agents is an enterprise AI agent platform from Sana Labs for creating and deploying agents that work with organizational knowledge, connect to business systems, and automate research, analysis, and recurring workflows.
❓ Frequently Asked Questions
What is SWE-agent?
SWE-agent is an open-source AI coding-agent framework created by researchers at Princeton University. It enables language models to navigate code repositories, modify files, run shell commands, and attempt to resolve software issues.
Is SWE-agent free and open source?
Yes. SWE-agent's source code is available on GitHub under the MIT license. Users still need suitable computing resources and, depending on the selected model, may incur API or model-hosting costs.
How does SWE-agent work?
SWE-agent gives a language model a specialized Agent-Computer Interface for viewing files, searching code, editing repositories, and executing commands. The agent iteratively investigates an issue, changes code, and checks its work.
Can SWE-agent fix GitHub issues automatically?
SWE-agent is designed to attempt fixes for GitHub issues in real repositories, but results vary by issue, model, configuration, and test coverage. Generated changes should be reviewed and tested before use.
What are the best SWE-agent alternatives?
Relevant alternatives include Devin for autonomous software-engineering tasks, GitHub Copilot and Cursor for AI-assisted coding, and Replit Agent for building and modifying applications through an agent workflow.