SWE-agent
Princeton's research coding agent — now points users to mini-swe-agent
Overview
SWE-agent is Princeton NLP's open-source agent that resolves GitHub issues via an LLM-driven Agent-Computer Interface. Actively maintained, but the team's current focus has shifted to its 100-line successor, mini-swe-agent.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 11 Jul 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
Use cases
What you can produce with SWE-agent
- Open-source (MIT), self-hosted agent — bring your own LLM API key
- Agent-Computer Interface (ACI) for structured repo navigation and editing
- Resolves GitHub issues via pull request-style patches
- Single-YAML configuration, fully documented
- EnIGMA variant for cybersecurity/CTF tasks
- Vision-model support for issues containing images
- Benchmarked on SWE-bench / SWE-bench Verified
ASEAN Perspective
SWE-agent in Southeast Asia
ASEAN-region availability and pricing notes coming soon. Drop the editorial team a note via /contact/ if you can supply local context (Singapore/Malaysia/Indonesia/Thailand/Vietnam).
SWE-agent matters for what it proved: that a purpose-built interface between an LLM and a codebase (their Agent-Computer Interface concept) beats bolting a model onto raw shell access. It's MIT-licensed, well-documented, configurable via a single YAML file, and includes EnIGMA, a variant tuned for cybersecurity capture-the-flag tasks. At 19,800+ GitHub stars and commits as recent as this month, it's still a maintained repository, not an abandoned research artifact.
But Princeton's own README now says most current development effort goes into mini-swe-agent, a stripped-down successor that scores over 74% on SWE-bench Verified in roughly 100 lines of Python — a large jump from SWE-agent's original ~12% resolve rate at its 2024 launch. If you're picking this up today, expect to be steered toward the newer project. This is a tool for researchers and advanced developers bringing their own model keys, not a turnkey product — there's no hosted service, UI, or support line.
What people say
SWE-agent came out of Princeton NLP (with Stanford collaborators) and was one of the first agents to show that giving an LLM a well-designed command interface — rather than raw bash — meaningfully improves its ability to autonomously fix real GitHub issues. Its original Hacker News launch in 2024 reported a 12% resolve rate on SWE-bench issues, which at the time was a notable result; commenters were split between excitement about the approach and skepticism about how representative the underlying benchmark issues were of real day-to-day engineering work.
The project has kept moving. It's currently at 19,800+ GitHub stars with 30 open issues and commits dated as recently as July 2026, so it isn't stalled. It supports vision-capable models (Claude Sonnet 4, GPT-4o) for issues that include screenshots, and EnIGMA extends the same agent framework to offensive-security and CTF-style tasks. A 2026 security research note flagged SWE-agent among several widely used coding agents that could be tricked into running destructive shell commands via bash text-rewriting tricks that slipped past command-approval guards — worth knowing if you're running it with any write access to production systems.
The more important story for anyone evaluating it today is the succession plan. Princeton's own team has shifted its primary development effort to mini-swe-agent, a roughly 100-line reimplementation that trades SWE-agent's configurability for radical simplicity, and which now scores over 74% on SWE-bench Verified — a huge jump from where the original project started, though much of that gain reflects underlying model improvements (GPT-4o-class to current frontier models) as much as the scaffold itself. mini-swe-agent already has 5,700+ stars of its own. SWE-agent remains usable and documented, but new adopters should read both repos before choosing, since the maintainers are explicit that mini-swe-agent is where the active work is going.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including SWE-agent's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with SWE-agent unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to SWE-agent directly →
Spotted something out of date? Suggest an update →
SWE-agent in the news
More in Code & Dev Tools
