swesmith

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 on this page — 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

swesmith on PyPI

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

Evidence: swesmith-0.0.9-py3-none-any.whl

Keywords: nlp, benchmark, code

Tags

More Artificial Intelligence packages

Further reading