SWE-agent
Open-source research agent that turns a GitHub issue into a patch using the LLM you choose
SWE-agent earned its reputation honestly, and if you are studying agent-computer interfaces, reproducing SWE-bench numbers, or need EnIGMA's CTF work in 0.7, the original still has value. For everyone else, the repository's own warning applies: most development effort has moved to mini-swe-agent, which the team says matches this project's performance while being much simpler. Pick SWE-agent when you want to read, fork, or benchmark the architecture, not when you want a coding agent you can rely on next quarter. Teams wanting a supported product should look at commercial coding agents instead.
Verified 1d ago · liveness 74/100 · cite: rightaichoice.com/tools/swe-agent
- Researchers studying autonomous program repair or agent-computer interfaces
- Academics reproducing or benchmarking SWE-bench results
- Security researchers needing EnIGMA mode for CTF and offensive work
- Developers comfortable with CLI and Docker who want to fork the agent loop
- Beginners who need a visual interface or IDE plugin
- Teams expecting vendor support or a maintenance guarantee
- Production workflows that must keep pace with every new model release
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip SWE-agent if you want a supported coding agent with a GUI and vendor backing rather than a research platform in maintenance mode — the maintainers themselves steer new work to mini-swe-agent.
The software is free under the MIT license, but you pay your LLM provider for every token the agent burns exploring a repo and retrying failed edits
SWE-agent costs $0 under the MIT license — there is no seat fee, subscription, or tier to climb. The real spend is your LLM API bill or local GPU time, which scales with how many issues you run. Against paid coding agents that bundle model access and support into a monthly subscription, SWE-agent is cheaper on paper but shifts the integration and maintenance work onto your team.
In short
SWE-agent — Open-source research agent that turns a GitHub issue into a patch using the LLM you choose. Best for Researchers studying autonomous program repair or agent-computer interfaces, Academics reproducing or benchmarking SWE-bench results, Security researchers needing EnIGMA mode for CTF and offensive work. Free to use.
What's new in SWE-agent
Checked yesterdayAcross the latest 2 updates: 1 launch and 1 news mention.
Mini-SWE-Agent achieves 65% on SWE-bench verified in 100 lines of python!
The maintainers announced mini-swe-agent, a much simpler agent they say matches SWE-agent's performance and which they recommend for new work going forward.
SWE-agent-LM-32b achieves open-weights SOTA on SWE-bench
SWE-agent's 32B language model reached state-of-the-art results among open-weight models on SWE-bench.
What people actually say about SWE-agent — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
30 mentions across 1 source (Hacker News) · researched Jul 3, 2026.
Average across the 1 source that answered — each source counts once, not each post.
- +Autonomous bug fixing from a GitHub issue URL.
- +Supports multiple LLM backends: OpenAI, Anthropic, local models.
- +Sandboxed Docker environment ensures safe execution.
- +Top SWE-bench scores at time of NeurIPS 2024 publication.
- +Free and open-source under MIT license.
- −Replaced by mini-swe-agent for simpler use cases.
- −Complex Docker-based setup and configuration.
- −No GUI; entirely command-line driven.
- −Documentation can feel academic, not user-friendly.
- −Autonomous mode may produce incorrect fixes without review.
- • LLM API usage costs (pay-per-token) not included
- • Compute and storage for Docker images
Viability Score
How well maintained and how widely used is SWE-agent? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- Autonomously fix GitHub issues from a plain-language issue description
- Open-source under the MIT license with no license fee or seat cost
- Use your choice of LLM: GPT-4o, Claude Sonnet 4, or local models via Ollama
- Run agent commands inside a Docker sandbox for isolation
- Generate patches and pull requests with proposed fixes
- Configure agent behavior through a single YAML file
- Log full trajectories for debugging and research analysis
- EnIGMA mode for offensive cybersecurity and CTF challenges
- Benchmark against SWE-bench lite, verified, and full
- Command-line interface (CLI), no graphical GUI
- Extensible tool set for file editing, shell commands, and custom tasks
- Support for competitive coding challenges
- Interactive commands and a summarizer for long contexts
- Try the agent in your browser via the online demo
About SWE-agent
SWE-agent is an MIT-licensed research project from Princeton and Stanford that points a language model of your choice at a GitHub issue and works toward a fix. The agent calls tools to explore a repository, edit files, and run shell commands inside a Docker sandbox, then produces a patch or pull request. Named backends in the project's own documentation include GPT-4o and Claude Sonnet 4, and you can also run local models through Ollama. Agent behavior is governed by a single YAML config file, trajectories are logged for debugging and analysis, and benchmarking harnesses cover SWE-bench lite, verified, and full. A separate EnIGMA mode targets offensive cybersecurity and capture-the-flag work; the maintainers ask that you use SWE-agent 0.7 while EnIGMA is updated for 1.0. Published results include SWE-agent 1.0 with Claude 3.7 reaching state of the art on SWE-bench full and verified, and SWE-agent-LM-32b taking open-weights SOTA on SWE-bench. The project is pitched at developers, security researchers, and academics studying AI-driven software engineering, not at teams wanting a managed product — it runs from a CLI with Docker and has no graphical interface. The dominant caveat is maintenance: the maintainers state that most current development effort is on mini-swe-agent, which they say matches SWE-agent's performance while being much simpler, and they recommend it for new work.
Behind the Verdict
SWE-agent is best understood as a research artifact with a real track record rather than a product you adopt. What it actually ships is an agent loop: a language model of your choice drives tool calls to open files, edit code, and run shell commands inside a Docker sandbox, and the run ends with a patch or pull request. That loop is governed by a single YAML file, which is the honest heart of the thing — it means you can change agent behavior by editing configuration rather than rewriting Python, and the trajectory logs let you inspect why the agent did what it did. The results are documented rather than implied: SWE-agent 1.0 with Claude 3.7 posted state of the art on SWE-bench full and verified, SWE-agent-LM-32b took open-weights SOTA on SWE-bench, and the work was presented at NeurIPS 2024. Benchmarking harnesses for SWE-bench lite, verified, and full are included, which is why academics keep reproducing results on it. Strengths: transparency, MIT licensing with no fee or seat cost, freedom to point it at GPT-4o, Claude Sonnet 4, or a local Ollama model, and EnIGMA mode for offensive cybersecurity and CTF challenges. Weaknesses: it is a CLI tool requiring Docker with no graphical interface, patch quality tracks the underlying model closely, and the maintainers themselves now steer new work to mini-swe-agent, which they say matches performance at a fraction of the complexity. Where it fits: labs and individual engineers who want to fork or measure an agent loop. Where it does not: teams that need vendor support, a visual interface, or an agent that keeps pace with new model releases without them doing the work.
Researching SWE-agent? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas SWE-agent actually fits — and what changes day-one when you adopt it.
You clone the repository, install it with Docker, write an API key into the environment, then run the SWE-bench harness against the lite split with a model you want to evaluate.
Outcome: You get comparable pass rates across models, full trajectory logs to inspect agent reasoning, and a YAML config you can change to test a modified tool set.
You point the agent at a real GitHub issue from the command line, let it explore the repository and run shell commands in the Docker sandbox, and review the patch it produces.
Outcome: You get a candidate pull request for a straightforward bug without writing the first draft yourself, and you keep full visibility into every step it took.
You check out SWE-agent 0.7, enable EnIGMA mode, and turn the agent loose on a capture-the-flag task rather than a repository fix.
Outcome: You get an agent configured for offensive cybersecurity challenges, with published benchmark results you can compare your runs against.
Use Cases
- Automatically generate pull requests for open GitHub issues
- Evaluate LLM performance on software engineering tasks using SWE-bench
- Simulate autonomous agents for cybersecurity red team exercises
- Practice competitive programming with AI co-pilots
- Prototype automated bug bounty solvers
- Conduct research on code generation and repair
Models Under the Hood
as of 2026-09-29
Limitations
- SWE-agent requires running locally or on a server with Docker, and it is a CLI tool with no graphical interface.
- Performance depends heavily on the underlying LLM; weak models may produce incorrect patches, and the tool is best for straightforward bugs rather than highly complex or context-dependent issues.
- EnIGMA mode for offensive cybersecurity currently requires SWE-agent 0.7 while it is updated for 1.0.
- The original tool is now in maintenance mode: the maintainers state that most current development effort goes to mini-swe-agent, which they say matches this project's performance while being much simpler, and they recommend it for new work.
as of 2026-10-07
Verification history
We have re-verified SWE-agent 9 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
Showing the 6 most recent of 9 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Plans compared
For each published SWE-agent tier: who it actually fits, and what it adds vs. the previous tier. Cross-reference the cost calculator above for projected annual outlay.
Free (open source)
$0
Ideal for
Researchers, academics, and developers who can run Docker locally and supply their own LLM API key or Ollama model
What this tier adds
Starting tier: MIT-licensed software at no cost, with benchmarking harnesses, trajectory logging, and EnIGMA mode included
Where the pricing makes sense
The company stage and team size where SWE-agent's pricing actually pencils out — and where peers do it cheaper.
SWE-agent costs $0 under the MIT license — there is no seat fee, subscription, or tier to climb. The real spend is your LLM API bill or local GPU time, which scales with how many issues you run. Against paid coding agents that bundle model access and support into a monthly subscription, SWE-agent is cheaper on paper but shifts the integration and maintenance work onto your team.
Setup time & first value
How long it actually takes to get something useful out of SWE-agent — broken out by persona, not the marketing-page minute.
Researchers with Docker experience: roughly 30 to 60 minutes from clone to a first benchmark or issue run, mostly Docker build time. Developers without Docker: budget an afternoon for the environment, an API key or local Ollama model, and one debugging pass on the YAML config. CTF users: add time to fetch the 0.7 checkout that EnIGMA currently requires.
Switching to or from SWE-agent
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From a manual issue triage workflow: point SWE-agent at open issues and review the patches it proposes instead of writing every first draft
- →From mini-swe-agent: the agent loop and YAML configuration model are the same family, so trajectories and configs carry over conceptually
- →From a commercial coding agent: replicate the loop locally when you need trajectory logs and a forkable architecture you can audit
- ↗To mini-swe-agent: the maintainers say it matches SWE-agent's performance while being much simpler, and they recommend it for new work
- ↗To a commercial coding agent: move here when you need a GUI, vendor support, and a tool that tracks new model releases for you
Integrations
Resources & Guides
Tutorials & Learning
SWE-Agent: AIエージェントがアプリやコードなどを自動でコーディング・構築！(オープンソース)
WorldofAI
AIエージェントがツールを使って自動的にコードを生成 - SWE-Agentチュートリアル（「Devin Clone」）
Matthew Berman
SWE-agent：将大模型转变为软件工程师Agent，可以修复 GitHub 项目中的错误和问题，SWE-bench 上SWE-agent 解决了 12.29% 的问题，接近devin
AIGCLINK
YouTube returned 6 videos for “SWE-agent”, and we withheld 3: 3 did not mention SWE-agent. Showing the 3 we can prove are about SWE-agent.
Official links
Tools that pair well with SWE-agent
Common stack mates teams adopt alongside SWE-agent, with the specific reason each pairing earns its keep.
Open Interpreter
Open Interpreter is an open-source terminal agent that turns plain-English requests into real file edits and shell commands on your machine.
Goose
Open-source AI agent for desktop, CLI, and API — automate code, research, and writing with your own LLM keys or subscriptions
DeerFlow
Open-source SuperAgent harness that researches, codes, and creates inside a persistent Docker sandbox — MIT licensed and self-hosted.
Featured Head-to-Head Comparisons
Swe Agent vs Locus Robotics
Locus Robotics and SWE-agent solve completely different problems: warehouse logistics vs. software bugs. Your choice depends on domain. If you need to physically move goods in a warehouse, Locus is proven with real-world clients; if you need to automate GitHub issue resolution, SWE-agent is free and open-source. There is no overlap—buyers should evaluate based on their operational need, not feature comparison.
Swe Agent vs Truleo
If you're in law enforcement needing to unearth leads from scattered data, Truleo's purpose-built suite (jail call analysis, BWC analysis, report writing) is unmatched. For developers automating bug fixes, SWE-agent's open-source flexibility and LLM-agnostic design make it a powerful free tool. Choose based on your domain: policing or programming.
Swe Agent vs Presto Voice
These tools serve completely different markets: Presto Voice is a commercial drive-thru automation platform for QSR chains, while SWE-agent is a free, open-source bug-fixing tool for developers. Choose Presto if you run a multi-location QSR and want to boost revenue via voice AI; choose SWE-agent if you manage open-source projects and need automated patch generation. There is no overlap.
Alternatives to SWE-agent
View all
Open Interpreter
Open Interpreter is an open-source terminal agent that turns plain-English requests into real file edits and shell commands on your machine.
Frequently Asked Questions
Categories
Best-of guides
Used SWE-agent? Help shape our editorial sentiment research.