SWE-agent/mini-swe-agent

A minimal AI coding agent written in approximately 100 lines of Python that operates exclusively through bash execution. It accepts natural language prompts and GitHub issues as inputs, returning shell commands, code modifications, or resolved issue outputs, but requires a connection to an external language model provider such as LiteLLM, OpenRouter, or Portkey to operate.

Project overview

The project addresses overly complex software engineering agents by providing a minimalist alternative that scores above 74% on SWE-bench verified without relying on custom tools or language model tool-calling interfaces.

Project type: AI Agent
Deployment: Refer to project documentation
License: MIT

Best for

Developers and researchers seeking a minimal command line agent for ai-coding and ai-agent-automation tasks.

Key capabilities

An AI agent that helps users solve tasks directly in the command line.
An AI agent designed to help resolve GitHub issues.
Operates without custom tools or LM tool-calling interfaces, relying exclusively on bash execution.

Limitations and risks

The agent does not have any tools other than bash.
Agent utility relies entirely on the language model utilizing the shell to its full potential rather than structured tool calling.

Getting started

The setup difficulty is rated easy because it installs directly via pip in a single command and provides pre-configured virtual environment execution via uvx. Users can run 'pip install uv && uvx mini-swe-agent' and then run 'mini' in the command line.

Evidence and sources

Open original GitHub project

AI Search

Find projects, verify facts, compare options, or turn a complex need into an actionable plan

Try a searchA click only fills the search box; you stay in control

Collection Scenario Project type License Deployment Resource form

SWE-agent/mini-swe-agent

Project overview

Best for

Key capabilities

Limitations and risks

Getting started

Evidence and sources

AI Search

Project Details