Tool overview
mini-SWE is listed under Programming AI tools.
What is mini-SWE?
mini-SWE-agent is a deliberately small software-engineering agent that solves GitHub issues and command-line tasks through a bash-oriented control loop. It supports local use, multiple sandbox environments, batch benchmarks, model providers, trajectories, and Python customization.
Best for
Developers and researchers wanting a small, inspectable coding-agent scaffold
Who is it for?
Decision note
Confirm model provider, token costs, sandbox type, architecture compatibility, network and credential isolation, benchmark needs, supported configuration, maintainer releases, and mandatory human review controls.
Key features
Minimal approximately one-hundred-line agent core
Interactive command-line interface
GitHub issue and repository task solving
Docker, Singularity, and Bubblewrap environments
Batch benchmark and trajectory tooling
Model access through multiple providers
Use cases
Solving scoped GitHub issues
Assisting with local command-line tasks
Running SWE-bench evaluations
Researching simple agent control loops
Pros
- Open-source MIT-licensed code
- Small architecture is easy to inspect
- Supports local and sandboxed workflows
Limitations
mini-SWE-agent can execute destructive commands, expose secrets, introduce vulnerable code, or produce incorrect patches. Run it in isolated environments, restrict credentials and networks, inspect trajectories, execute tests, and require human code review.
Pricing details
Billing options
Pricing note
mini-SWE-agent is free, open-source software under the MIT license. Users supply their own model-provider access and pay any resulting inference, compute, container, storage, or infrastructure costs directly.
Supported programming languages
- English
Integrations
LiteLLM
OpenRouter
Portkey
Docker
SWE-bench
GitHub
Please log in to join the discussion.