
SWE-agent
Not yet source-checkedOpen-source autonomous agent framework that lets LLMs fix GitHub issues and find security vulnerabilities by using a custom agent-computer interface.
Machine-drafted from the vendor's site · not yet source-checkedProfile updated Link checked How we rate
In short
SWE-agent is a free, open-source framework that lets LLMs fix GitHub issues and find vulnerabilities. It is a hackable research baseline, not a hosted product.
Pick SWE-agent if you're a researcher or engineer who wants a hackable, reproducible coding-agent baseline you can wire to any LLM.
Skip it if you want a plug-and-play hosted coding assistant - reach for Cursor, Devin, or Cline instead.
SWE-agent is an open-source autonomous coding agent built by researchers at Princeton and Stanford that turns a language model into a software engineer. It gives the LLM a tightly-scoped agent-computer interface (ACI) for navigating repos, editing files, running tests, and iterating on fixes, and it was one of the first agents to post serious numbers on SWE-bench - including state-of-the-art results among open-source projects when it launched. A sibling mode, EnIGMA, retargets the same loop at offensive-security CTF challenges.
It's aimed at researchers and engineers who want a hackable agent rather than a polished SaaS. Configuration is a single YAML file, you bring your own API key (GPT-4o, Claude Sonnet, DeepSeek, or local models via LiteLLM), and you run it yourself - there's no hosted product, no pricing page, no signup. The team has explicitly moved focus to mini-swe-agent, a 100-line successor, and labels the original repo as maintenance-only, so treat SWE-agent as a stable research baseline rather than an actively-developed product.
It integrates with GitHub issues, Docker-sandboxed execution, and the SWE-bench evaluation harness, and is widely cited as a reference implementation for tool-using coding agents. Caveat: LLM API costs are on you, and on real-world tasks they add up quickly.
SWE-agent is a foundational piece of the open coding-agent canon and still worth studying, but the team has moved on to mini-swe-agent and says so on the front page. Use it as a research reference or a starting fork; for daily-driver coding, pick something actively maintained.
— The AI Tool Bible editorial team
Pros
- ✅ Open-source under MIT with a strong research pedigree (Princeton/Stanford)
- ✅ Model-agnostic via LiteLLM - swap GPT-4o, Claude, or local models freely
- ✅ Reproducible SWE-bench harness makes it a credible baseline for agent research
- ✅ EnIGMA mode extends the same loop to CTF-style security tasks
Cons
- ⚠️ Officially in maintenance mode; team now recommends mini-swe-agent
- ⚠️ No hosted product, GUI, or managed service - CLI and YAML only
- ⚠️ LLM API costs on long-horizon tasks can be significant
- ⚠️ Setup requires Docker and comfort with Python tooling
Facts
- Platforms
- cli
- Open source
- Yes (license: MIT)
- GitHub stars
- 9,575 (checked Nov 13, 2026)
- Last GitHub push
- Nov 9, 2026
- First commit
- Jun 2025
- Model used
- Multi-model (GPT-4o, Claude Sonnet, DeepSeek, local via LiteLLM)
- Package downloads (30 d)
- 14,060,333 · pypi mini-swe-agent · checked Nov 22, 2026
- Hacker News
- 4 stories · top 6 points · last May 28, 2024 · checked Nov 22, 2026
- Earliest Wayback capture
- 2024-11-04 (earliest archived capture of the site, not a launch date)
Where this tool sits
Use cases
Frequently asked
- How much does SWE-agent cost?
- SWE-agent is free and open-source. However, you must pay your own LLM API costs, which can add up quickly on real-world tasks. There is no subscription or hosted pricing.
- What LLMs does SWE-agent support?
- It supports multiple models including GPT-4o, Claude Sonnet, and DeepSeek. You can also use local models via LiteLLM. You bring your own API key and configure it via a single YAML file.
- Is SWE-agent a hosted service?
- No, there is no hosted product, pricing page, or signup. You run it yourself using Docker-sandboxed execution. It is designed as a hackable, reproducible baseline for researchers and engineers, not a plug-and-play SaaS.
- What are the main use cases?
- It is best for fixing GitHub issues, autonomous coding, SWE-bench evaluations, and CTF security challenges. It provides a custom agent-computer interface for navigating repos, editing files, and running tests.
- Should I use SWE-agent or a tool like Cursor?
- Skip SWE-agent if you want a plug-and-play hosted coding assistant; reach for Cursor, Devin, or Cline instead. Choose SWE-agent if you need a hackable, open-source baseline you can wire to any LLM for research.