mini-SWE-agent

Categories: Coding & Developer Tools, Automation / Agents, Research | Pricing: Free | Official Website ↗

mini-SWE-agent is a minimalist, 100-line AI agent designed for software engineering tasks, capable of resolving issues using only bash.

mini-SWE-agent is a simplified AI agent for software engineering, built by the Princeton & Stanford team behind SWE-bench and SWE-agent. It is designed to be minimal, consisting of only 100 lines of Python code, plus 100 lines for environment, model, and script components. Despite its simplicity, it achieves over 74% on the SWE-bench verified benchmark, making it a performant tool for research and development. The agent operates without specialized tools, relying solely on bash commands, which allows it to work with virtually any language model and simplifies sandboxed execution. Its linear history and independent action execution via `subprocess.run` enhance debugging, fine-tuning, and scalability. mini-SWE-agent supports various environments like local, Docker, Singularity, and more, and is compatible with multiple model providers through Litellm, OpenRouter, and Portkey.

Key Features

100-line Python core
Bash-only tool usage
Linear action history
Independent action execution via subprocess.run
Support for local, Docker, Singularity, Bubblewrap, Contree environments
Compatibility with Litellm, OpenRouter, Portkey models
Python bindings for agent control
Trajectory browser CLI

Pros

Extremely simple and minimal codebase (100 lines)
High performance on SWE-bench (>74%)
Supports any language model due to bash-only tool use
Simplified sandboxing and scalability
Linear history aids debugging and fine-tuning
Widely adopted by major tech companies and universities

Cons

Limited to bash for all operations, no custom tools
Requires familiarity with command-line usage
May not be suitable for complex, multi-tool agent experiments
Focuses on the LM rather than agent scaffolding
Documentation might require technical understanding

Use Cases

Automated software bug fixing
Code generation and refactoring
Benchmarking language models on coding tasks
Developing and testing AI agents in sandboxed environments
Research into minimalist AI agent architectures

Best For

AI agent researchers
Developers needing a simple, stable coding agent
Teams performing fine-tuning or reinforcement learning on agents
Users evaluating language model performance on coding tasks
Organizations requiring deployable and compatible AI agents

Integrations: Litellm, OpenRouter, Portkey, Docker, Singularity, Bubblewrap, Contree

Platforms: Web, Linux, macOS, Windows

Watch demo on YouTube ↗

View full mini-SWE-agent profile on Tools-Radar | Browse Coding & Developer Tools tools | Alternatives to mini-SWE-agent

Tools-Radar is a free directory of 10,000+ AI tools — discover, compare, and choose the right AI software for your needs. Visit tools-radar.com