Our take
- Activity
- 64
- Community
- 47
- Issues
- 58
- Pull requests
- 36
+1,059 stars in the last 30 days
- Releases more often than 83% of the projects we track
Project history
Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.
Stars
9,615Forks
1,320Change since Nov 14, 2026: +20 stars·+3 forks
Overview
mini-swe-agent is a Python agent for software engineering work, designed to solve tasks such as GitHub issues through a language model that acts using bash. It aims to keep the agent scaffold small and understandable rather than relying on a large tool framework or complex configuration.
Its linear message history and independent command execution can make behavior easier to inspect and adapt. The project is intended both as a practical command-line tool and as a simple baseline for evaluating or training models without tightly coupling experiments to a complex agent scaffold.
Key Features
- Provides a command-line interface and Python bindings.
- Uses bash as its agent interface rather than a collection of custom tools.
- Appends each step to a linear message history that can be inspected.
- Runs actions independently with
subprocess.run, supporting execution in different environments. - Supports local, Docker/Podman, Singularity/Apptainer, bubblewrap, and contree environments, according to the README.
- Supports models through integrations including LiteLLM, OpenRouter, and Portkey.
- Includes batch inference for SWE-bench and a trajectory browser.
Use Cases
- Developers can use the CLI to ask an agent to investigate or modify a local codebase.
- Research teams can compare language models using a compact agent scaffold with a linear trajectory.
- Engineers can run software engineering tasks in local or sandboxed environments without building custom agent tools.
- Developers extending agent workflows can use the Python bindings and adapt the environment or model integration.
Getting Started
Install and launch the CLI:
pip install mini-swe-agent
mini
See the README and documentation for configuration, environment setup, and Python usage.
Alternatives
- grok-build: Grok Build is a Rust agent with a full-screen terminal UI and web search, while mini-swe-agent is a compact Python agent built around a simple bash loop.
- gpt-engineer: GPT Engineer focuses on generating and iterating on code from natural-language specifications, while mini-swe-agent targets software engineering tasks such as GitHub issues.
- opencode: OpenCode offers build and read-only plan agents plus terminal and desktop interfaces, while mini-swe-agent is a simpler Python CLI agent.
- pi: Pi is a TypeScript toolkit and CLI with a unified API for multiple model providers, while mini-swe-agent is a compact Python agent centered on bash.
| Project | Language | License | Stars | Status |
|---|---|---|---|---|
| mini-swe-agent | Python | MIT | 8.2k | Not checked yet |
| grok-build | Rust | Apache-2.0 | 27.3k | New |
| gpt-engineer | Python | MIT | 55.1k | Archived |
| opencode | TypeScript | MIT | 212k | Active |
| Pi | TypeScript | MIT | 117k | Active |
Considerations
- The agent relies on a language model, so users need to configure an appropriate model provider or endpoint.
- It uses bash for actions. Commands can modify the working environment, so choose an execution environment and permissions appropriate to the task.
- The README identifies the repository as v2 and links to a migration guide; users of v1 should review that guide before upgrading.
- The project emphasizes a small scaffold, so users who need specialized built-in tools or configurable history processing may prefer a more extensible agent framework.
Found this useful?
Share it with someone who would like mini-swe-agent.