mini-SWE-agent
Minimal open-source coding agent from the SWE-bench and SWE-agent team that solves GitHub issues or terminal tasks using only bash commands, with any language model.
About mini-SWE-agent
Larger coding-agent scaffolds with custom tools and stateful shells are hard to understand, sandbox and reuse as a clean baseline for benchmarks, fine-tuning or daily command-line work.
Who it’s for
- Researchers running SWE-bench-style evaluations or training models on agent trajectories
- Developers who want a small, hackable command-line coding agent they can read and extend
When to consider it
Consider mini-SWE-agent when you want a coding agent whose core loop is about 100 lines of Python, works with any model via LiteLLM, OpenRouter or Portkey, and runs actions locally or in Docker, Podman, Singularity/Apptainer or Bubblewrap sandboxes.
Tradeoffs & limitations
- MIT-licensed and self-run: you install it with pip, uv or pipx and supply your own model API keys or local models; there is no hosted service.
- The agent has no tools other than bash and runs each command as an independent subprocess, so features such as opening pull requests are left to the model, and sandboxing depends on the environment you configure.
- It supersedes SWE-agent, which its team has put in maintenance-only mode.
SOTA overview · Documentation-based assessment · Sources & review method
Updates
No updates shared yet.
Discussion
Newest firstAsk a question or share how you use mini-SWE-agent.
Keep it helpful. Community rules
Loading discussion…
Sources & review method
Documentation-based assessment · Nov 1, 2026 · Prepared with AI assistance; not a hands-on benchmark.
- mini-SWE-agent documentation
- mini-SWE-agent quick start
- mini-SWE-agent on GitHub
- SWE-agent documentation (maintenance-mode notice)
Checked by SOTA · AI-assisted documentation review. Selection advice is our assessment; verify current requirements for your deployment.
Import history & original evidence
mini-SWE-agent official documentation
Minimal open-source coding agent from the SWE-bench and SWE-agent team that solves GitHub issues or terminal tasks using only bash commands, with any language model.
Based on official pages and announcements checked on 2026-11-01. No hands-on test, performance benchmark or popularity ranking is claimed.