inspect-swe

What it is and what it does

Inspect SWE is a collection of software engineering agents designed to work within the Inspect AI framework. It provides tooling for automated evaluation of code generation, testing, and other software engineering tasks. The package is built on top of Inspect AI and integrates with its agent protocol, allowing you to define and run evaluations against AI models or code generation systems.

The package depends on Inspect AI as its core runtime, along with standard utilities like pydantic for validation, httpx for HTTP operations, and asyncio helpers. It's actively maintained and recently released, with a focus on sandbox-based testing—most tests require Docker or Kubernetes to be available. The MIT license and low install friction make it straightforward to add to existing projects.

Use it for:

• Evaluate code generation models against software engineering benchmarks and tasks
• Build automated testing pipelines for AI-assisted coding systems
• Run reproducible evaluations of code quality and correctness across multiple agents
• Integrate code evaluation into CI/CD workflows using Inspect AI's framework
• Benchmark different software engineering agents on standardized tasks

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Inspect SWE provides a suite of software engineering agents built on Inspect AI, enabling automated evaluation and testing of code generation and engineering tasks.

Yes, if you are working with Inspect AI and need to evaluate code generation or software engineering agents. The package is actively maintained, has low install friction, carries a permissive MIT license, and has no known vulnerabilities. The main consideration is that most tests require a sandbox environment (Docker or Kubernetes), so verify that your setup can provide one before relying on the full test suite.

Install

inspect-swe on PyPI

pip install inspect-swe

uv add inspect-swe

poetry add inspect-swe

Installing inspect-swe

Before you install

Low friction install with a pure Python wheel. Active maintenance—last commit 5 days ago—and recent release cycle (5 days since latest version). Depends on 10 runtime packages including inspect-ai, pydantic, and httpx; all are well-established.

License in practice

MIT License permits commercial and private use with minimal restrictions, making it suitable for most projects without legal friction.

Quickstart

pip install inspect-swe

from inspect_swe import <agent_class>
agent = <agent_class>()
# Use agent for code evaluation tasks

Requires Python 3.10 or later. Most tests depend on a valid sandbox (Docker or Kubernetes) being available in your environment.

Verify before relying

• Which specific agent classes are available and what tasks each one is designed to handle
• Whether the package works with specific LLM providers or if additional configuration is required
• Performance characteristics and typical evaluation runtime for common benchmarks

Package facts

Evidence: inspect_swe-0.2.70-py3-none-any.whl

Tags

More Artificial Intelligence packages

Further reading