mini-swe-agent

mini-swe-agent

Open Source(MIT)
Verified

The 100-line AI agent that solves GitHub issues and assists in your command line.

Global
Self-hostable
Function calling
Public API
Free Forever
+2
Autonomous
Last updated: Oct 7, 2026
Compiled and verified by Intelloro · Ref INT-6QVZ9E
Open Source
Visit Site8.2k
More
mini-swe-agent.com

Quick Answer

AI

mini-SWE-agent is an open-source AI agent framework for researchers and developers building baseline AI systems. It achieves high performance on SWE-bench, solving GitHub issues and automating command-line tasks. Best for teams doing fine-tuning or reinforcement learning on agent scaffolds. Open-source.

About mini-swe-agentAI

mini-SWE-agent is an open-source AI agent framework built by Princeton and Stanford, designed to solve GitHub issues and assist with command-line tasks. It features a minimal Python codebase, high performance on SWE-bench, and supports various local and containerized environments. The agent is open-source and compatible with many LLM backends via LiteLLM, OpenRouter, and Portkey.

AI-extracted from public sources. May contain errors. Methodology · Report an error
What Makes mini-swe-agent UniqueAI

mini-SWE-agent is radically simpler than other agents, including its predecessor SWE-agent, by relying solely on bash for tool use and maintaining a completely linear history for debugging and fine-tuning. This minimal design allows it to run with virtually any language model and execute actions independently in sandboxed environments.

Decision Intelligence

Best For

AI
Researchers building baseline AI agent systems
Developers seeking a simple and hackable coding agent
Teams doing fine-tuning or reinforcement learning on agent scaffolds
Users who need a quick command-line tool for local development
Pricing Overview

Type

Open Source

Starting Price

$0/mo

Free Trial

No

Credit Card

Not Required

Free Plan

Yes

Dimension Scores

· 4.9 avg

How good it is, across universal quality axes

Ease of Use1/10
Customization7/10
Support8/10
Integration4/10
Compliance & Data Protection1/10
Performance8/10
Planning Quality5/10
AI-scored from up to 37 public sources, reviewed by a person·Medium confidenceHow we score

Task Scores

· 5.1 avg

What it can do, per category it works in · showing 6 of 9 tasks scoring 5+

Bug Detection7/10
Context Understanding7/10
Refactoring6/10
Multi Language Support6/10
Code Explanation5/10
Test Generation5/10

Scores are AI-estimated from publicly available data — not an independent test or a verified user rating. How we rank →

Capabilities
Code
Agent
Categories
Code Assistant
Automation Agents
LimitationsAI
  • -Does not have any tools other than bash
  • -Does not use the tool-calling interface of LLMs
  • -Requires Python knowledge for advanced customization
Integrations & Connections
4 connections
m
S
SWE-agent
SWE-agent

Open-source autonomous agent for GitHub issues & coding challenges

LI
LiteLLMNative

Supports all models via LiteLLM

OP
OpenRouterNative

Supports all models via OpenRouter

PO
PortkeyNative

Supports all models via Portkey

Auto-updated

Fit Score

Complexity

Advanced

Company Size

Solo
Small
Medium
Enterprise

Team Size

Individual
Small Team
Large Team

Skill Level

Expert
Features
Minimal 100-line Python codebase
Bash-only tool execution
Linear history for debugging and fine-tuning
Independent action execution via subprocess.run
Supports local, Docker, Singularity, Bubblewrap, and Contree environments
Compatible with all models via LiteLLM, OpenRouter, and Portkey
Scores >74% on SWE-bench verified benchmark
Python bindings for programmatic control
Technical Capabilities
SDK Available
MCP Compatible (unknown)
Multi-Agent (unknown)
Human-in-Loop
Team Features
Webhook Support (unknown)
Has Demo
Function Calling
Sandbox Mode (unknown)
Audit Logs
No Signup (unknown)
Works Offline (unknown)
Requires Host App
Framework
MCP Server

Deployment

Self Hosted
Developer Access
API
Public
HITL mode:
Approve
Compliance, Privacy & Trust
GDPR: No info available
SOC 2: No info available
HIPAA: No info available
Trains on user data: Unknown
Human review:
Recommended
Tech Specs
Response time (est.):
Fast

The Model(s) Behind It

Model details — extracted from public vendor sources by our pipeline (AI extraction plus a source check) or admin-entered. Shown as claimed by the vendor; not independently audited by Intelloro.

Powered By
Connects to 576+ models
Bring your own model — use your own API key or endpoint

Underlying Model Benchmarks

Third-party leaderboard scores for the AI model(s) powering this — including price & speed. These describe the underlying model, not the product's own metrics.

llm-coding

SWE-bench Verified
Princeton + OpenAI 2024 · SWE-bench
76.8
77th pct

Best submission: "Claude 4.5 Opus (high)" (2026-02-17)

11/15/2026·source

Scores from official leaderboards or vendor-reported. Percentile is cross-benchmark normalized when populated by an aggregator. See methodology.

Supported Tools
Bash Tool (Terminal)
Pricing Plans
Free
Free

    Looking for more mini-swe-agent alternatives?

    Explore our complete list of alternatives with filters for free, cheaper, and higher-rated options.

    View All Alternatives

    User Reviews

    No reviews yet. Be the first to review!

    Who It's For
    4 roles
    Software DevelopersComputer ProgrammersWeb Developers+1 more
    Platform, Content & Languages
    2 content types
    1 languages
    CodeTextEnglish+4 more

    Content Types

    Code
    Text

    Accepts

    Code
    Text

    Produces

    Code
    Text
    Languages: English

    Frequently Asked Questions

    What can mini-swe-agent do?

    mini-swe-agent can: Minimal 100-line Python codebase, Bash-only tool execution, Linear history for debugging and fine-tuning, Independent action execution via subprocess.run. Best for: Researchers building baseline AI agent systems.

    How much does mini-swe-agent cost?

    mini-swe-agent is open source and free to self-host. Cloud hosting options may have separate pricing.

    Is mini-swe-agent free?

    Yes, mini-swe-agent is open source and free to self-host. Managed hosting may carry separate costs.

    What makes mini-swe-agent unique?

    mini-SWE-agent is radically simpler than other agents, including its predecessor SWE-agent, by relying solely on bash for tool use and maintaining a completely linear history for debugging and fine-tuning. This minimal design allows it to run with virtually any language model and execute actions independently in sandboxed environments.

    Does mini-swe-agent have an API?

    Yes, mini-swe-agent offers a public API for developers to integrate its capabilities into custom applications and workflows.

    What platforms does mini-swe-agent support?

    mini-swe-agent is available on CLI and API.

    What languages does mini-swe-agent support?

    mini-swe-agent supports English.

    Who is mini-swe-agent for?

    mini-swe-agent is designed for Researchers building baseline AI agent systems; Developers seeking a simple and hackable coding agent; Teams doing fine-tuning or reinforcement learning on agent scaffolds.

    Trust Score

    60/100

    Good

    Based on 7 data signals

    Evidence: 4 of 12 expected facts verified · highest possible score 60

    Evidence strength: Strong

    Reflects publicly available data, not an independent audit

    Compliance & Verified38/38
    VerifiedYes
    Open-source licenseMIT
    Human ReviewYes
    Human-in-the-loopYes
    HITL TypeYes
    Reliability0/10

    No published data

    Market Proof10/10
    Notable CustomersMeta, NVIDIA, Essential AI
    Company17/17
    Company ListedYes
    Health—

    Not yet checked

    AI-Assisted Data

    Some details are AI-generated estimates — verify with the vendor.

    AI-generated estimate: Best For, Description, Limitations, Quick Answer, Tagline, What Makes Unique

    How we source data

    AI Governance & Regulatory Disclosures

    Legally-sensitive fields — extracted from public vendor sources by our pipeline (AI extraction plus a source check) or admin-entered. Shown as claimed by the vendor; not independently audited by Intelloro.

    EU AI Act Risk Tier
    Unclassified

    Switching & Migration

    Migration Difficulty
    Easy to Switch
    Support:
    GitHub
    Slack
    Community
    Quick Stats
    Agent TypeAutonomous
    MemoryShort Term
    PlanningBasic
    PlatformsCLI, API
    VerifiedOct 2026
    Company & Links
    Company NamePrinceton University & Stanford University
    Based inUnited States
    Last major updateJan 2024
    DocumentationChangelog

      We value your privacy

      We use cookies to enhance your browsing experience, analyze site traffic, and personalize content. Privacy Policy•Cookie Policy