---
id: swesmith
version: "0.0.9"
license: MIT License Copyright (c) 2025 John Yang Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in… (full text in the JSON record)
license_treatment: permissive
maintenance: active
---

# swesmith — The official SWE-smith package - A toolkit for generating software engineering training data at scale.

License: permissive · Maintenance: active · Downloads: 5.0M/mo

## What it is and what it does

SWE-smith is a toolkit for generating software engineering training datasets at scale. It converts GitHub repositories into executable environments and synthesizes task instances—such as file localization or program repair tasks—that can be used to train language models to perform software engineering work. The package manages Docker-based execution environments, generates synthetic tasks, filters them by test coverage, and produces issue descriptions for training.

The toolkit is designed for researchers and practitioners building datasets for software engineering agents. Installation from source is required to build custom datasets; the package depends on Docker and is tested only on Ubuntu 22.04.4 LTS, with explicit non-support for Windows and macOS.

Use it for:

- Generate custom software engineering benchmarks from any GitHub repository by creating execution environments and synthesizing task instances.
- Fine-tune language models on software engineering tasks using the pre-built dataset of 52k task instances.
- Create reproducible training data for software engineering agents with controlled task generation and test-based filtering.
- Build reinforcement learning datasets for code-based models using the container infrastructure.
- Evaluate software engineering agents on diverse repositories by synthesizing and executing task instances in isolated Docker environments.

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

SWE-smith generates large-scale software engineering training datasets by synthesizing task instances from GitHub repositories and managing their execution environments via Docker.

Yes, if you are training software engineering agents or building custom code benchmarks and have Docker and Ubuntu 22.04.4 LTS available. The toolkit is actively maintained, has low install friction, and provides a complete pipeline from repository to training dataset. Not suitable for Windows or macOS users, and requires significant infrastructure to use beyond pre-built resources.

## Install

pip install swesmith
uv add swesmith
poetry add swesmith

## Installing swesmith

Before you install: Low install friction with a pure Python wheel. Active maintenance (last commit 2026-08-10) and recent release cycle. Requires Docker and Ubuntu 22.04.4 LTS; Windows and macOS are explicitly unsupported.

License in practice: MIT License permits commercial and private use, modification, and redistribution with minimal restrictions. No warranty or liability protection for users.

Quickstart:

pip install swesmith
from swesmith.profiles import registry

Docker must be installed and running. Package requires Ubuntu 22.04.4 LTS; Windows and macOS are not supported. Python 3.10 or 3.11 required.

Verify before relying:

- What runtime dependencies are required beyond those listed in the fact sheet
- Specific Docker version or resource requirements for typical workflows
- Whether the package can be used on non-Ubuntu Linux distributions

## Package facts

- License: MIT License Copyright (c) 2025 John Yang Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in… (full text in the JSON record) (permissive)
- Python support: supports_current
- Install friction: low
- Maintenance: active
- Downloads: 5.0M/month (top 5,000 on PyPI)
- Known vulnerabilities: none known

## Tags

software engineering training data generation, github repository to benchmark dataset, swe-agent training toolkit, code task synthesis and harness, docker-based code execution environments, code-generation, benchmark-dataset, agent-training

[View on SkillFed](https://skillfed.io/packages/swesmith) · [View on PyPI](https://pypi.org/project/swesmith/)